Intelligent business process analysis method and system for multi-modal data

Abstract

The application provides a kind of intelligent business process analysis method and system for multi-modal data, it is related to enterprise operation management technical field, first, the multi-modal process data of business process is collected and process node correlation matrix is generated, then dynamic evolution correlation network is generated based on process node correlation matrix and real-time running state. Then the dynamic evolution correlation network is processed for bottleneck gene positioning, and the bottleneck gene features are extracted and input into the simulation adaptation scenario for dynamic deduction optimization to generate adaptive optimization parameters. Finally, based on the adaptive optimization parameters and the real-time state of the dynamic evolution correlation network, a dynamic adaptive business process optimization scheme is generated. The application can comprehensively and accurately analyze the business process, capture dynamic changes in real time, effectively locate bottlenecks and generate scientific optimization schemes, improve business process efficiency and reduce operating costs.

Description

Technical Field

[0001] This invention relates to the field of enterprise operation management technology, and more specifically, to an intelligent business process analysis method and system for multimodal data. Background Technology

[0002] In today's digital enterprise operating environment, the efficiency and stability of business processes are crucial. During operation, business processes generate a large amount of complex multimodal data, which covers multiple aspects such as system logs, user behavior, document flow, and collaborative interactions. These data collectively reflect the actual execution of the business processes.

[0003] However, existing business process analysis methods have significant shortcomings. On the one hand, traditional methods often focus only on a single type of data, such as analyzing system logs to understand system operation status or evaluating user experience solely based on user behavior data, failing to comprehensively grasp the overall state of the business process. On the other hand, there is a lack of effective analytical tools for the relationships and dynamic changes between nodes in the business process. Business processes are not static but evolve dynamically with time, business needs, and resource availability. However, existing technologies struggle to capture these dynamic changes in real time, making it difficult to promptly identify potential problems in the business process, such as poor data flow between nodes or unreasonable resource allocation. Consequently, it is difficult to provide targeted optimization solutions, leading to inefficiency and increased costs in the business process. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an intelligent business process analysis method for multimodal data, the method comprising:

[0005] Collect multimodal process data generated during the operation of business processes, perform process node link mapping processing on the multimodal process data, and generate a process node association matrix. The multimodal process data includes system log data, user operation behavior data, document flow data, and collaborative interaction data. The process node association matrix includes the connection relationship, data flow intensity, and resource consumption association of each process node.

[0006] Based on the process node association matrix and the real-time running status of the business process, the process node association network is dynamically evolved to generate a dynamically evolving association network. The dynamically evolving association network includes the dynamic change trajectory of node association strength, link evolution trend and network structure adaptability characteristics.

[0007] Bottleneck gene localization processing is performed on the dynamic evolutionary association network to extract bottleneck gene features in the process node association. The bottleneck gene features include node association obstruction features, resource allocation imbalance features, and link evolution stagnation features.

[0008] A simulation adaptation scenario is constructed, and the bottleneck gene features are input into the simulation adaptation scenario for dynamic deduction and optimization to generate adaptation optimization parameters. The adaptation optimization parameters include node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters.

[0009] Based on the adaptation optimization parameters, and in conjunction with the real-time status of the dynamically evolving network, a dynamically adaptable business process optimization scheme is generated. The dynamically adaptable business process optimization scheme includes process node association optimization strategies, resource configuration adjustment strategies, and link evolution guidance strategies.

[0010] In another aspect, embodiments of the present invention also provide an intelligent business process analysis system for multimodal data, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this embodiment of the invention collects multimodal process data during business process operation and performs process node link mapping processing to generate a process node association matrix containing the connection relationships of each process node, data flow intensity, and resource consumption correlation, which can accurately present the static structural characteristics of the business process. Based on this process node association matrix and the real-time running status of the business process, the process node association network is dynamically evolved to generate a dynamically evolving association network. This network can capture the dynamic changes of the business process in real time, including the dynamic change trajectory of node association strength, link evolution trends, and network structure adaptability characteristics, extending the analysis of the business process from static to dynamic, making it more closely aligned with actual business scenarios. Bottleneck gene localization processing is performed on the dynamically evolving association network, and the extracted bottleneck gene features can accurately identify the key problems in the business process. By constructing a simulation adaptation scenario and inputting the bottleneck gene features for dynamic inference and optimization, the generated adaptation optimization parameters provide a scientific basis for business process optimization. Ultimately, a dynamic adaptive business process optimization scheme is generated based on the adaptive optimization parameters and the real-time status of the dynamically evolving associated network. This scheme includes process node association optimization strategies, resource configuration adjustment strategies, and link evolution guidance strategies. It can comprehensively and accurately optimize business processes from multiple dimensions, effectively improving business process efficiency and reducing operating costs. Attached Figure Description

[0012] Figure 1This is a schematic diagram of the execution flow of the intelligent business process analysis method for multimodal data provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of the hardware architecture of the intelligent business process analysis system for multimodal data provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an intelligent business process analysis method for multimodal data provided in one embodiment of the present invention. The following is a detailed description of this intelligent business process analysis method for multimodal data.

[0015] Step S110: Collect multimodal process data generated during the operation of the business process, perform process node link mapping processing on the multimodal process data, and generate a process node association matrix. The multimodal process data includes system log data, user operation behavior data, document flow data, and collaborative interaction data. The process node association matrix includes the connection relationship, data flow intensity, and resource consumption association of each process node.

[0016] This embodiment uses an enterprise procurement business process as an application scenario for illustration. In this scenario, the collection of multimodal process data covers multiple stages, including purchase requisition, supplier selection, contract signing, order placement, logistics tracking, warehousing and acceptance, and financial payment. System log data includes operation records of each stage in the enterprise ERP system, such as the creation time of the purchase requisition and the processing time of approval nodes; user operation behavior data includes the operation sequence of purchasing specialists in the system, such as clicking, filling, and submitting; document flow data involves the transmission records of documents such as purchase requisitions, supplier qualification documents, contract texts, and acceptance reports; collaborative interaction data covers communication records between the purchasing department and suppliers, finance department, and warehousing department, such as email exchanges and instant messaging messages. The process node link mapping processing of the above multimodal process data is to correspond the above scattered data with the various specific execution stages in the procurement business process, thereby constructing a matrix that can reflect the relationship between each stage.

[0017] Step S111: Extract the process node identifiers corresponding to each type of data in the multimodal process data. The process node identifier is a unique identifier that represents an independent execution link in the business process, and each process node identifier corresponds to a specific business execution action.

[0018] In the enterprise procurement business process, the process node identifiers corresponding to various types of data need to be accurately extracted. For example, in system log data, the process node identifier corresponding to the purchase requisition stage is "CG-SQ-001," which uniquely corresponds to the specific business action of purchasing a purchase requisition; the supplier selection stage corresponds to "CG-XZ-002," representing the action of selecting a supplier. From user operation behavior data, when a procurement specialist performs the "submit purchase requisition" operation, this operation corresponds to the process node identifier "CG-SQ-001"; when performing the "screen supplier" operation, it corresponds to "CG-XZ-002." In document flow data, the flow of purchase requisition forms corresponds to "CG-SQ-001," and the flow of contract texts corresponds to the process node identifier "CG-HT-003" of the contract signing stage. In collaborative interaction data, communication between the procurement department and suppliers regarding quotations corresponds to the supplier selection stage "CG-XZ-002," and confirmation of payment terms with the finance department corresponds to the finance payment stage "CG-FK-006." By analyzing multimodal process data, a correspondence is established between different data types and corresponding process node identifiers to ensure that each process node identifier can be accurately associated with a specific business execution action.

[0019] Step S112: Analyze the execution time series corresponding to the process node identifier in the system log data, and generate the time distribution characteristics of each process node identifier. The time distribution characteristics include the execution start time, execution duration, and execution interval period.

[0020] For each process node identifier extracted from the system log data, its execution time series is analyzed to generate time distribution characteristics. Taking the process node identifier "CG-SQ-001" (purchase application) as an example, the time records of multiple executions of this node are obtained from the system log. The execution start time is the time when each purchase application is created, such as one execution start time being 9:15 AM on March 1, 2024, and another being 10:03 AM on March 5, 2024. The execution duration is the time elapsed from creation to submission for approval. For example, one purchase application starts at 9:15 AM and is submitted at 9:40 AM, with an execution duration of 25 minutes; another starts at 10:03 AM and is submitted at 10:30 AM, with an execution duration of 27 minutes. The execution interval is the time interval between two adjacent purchase applications. For example, the execution interval between two purchase applications on March 1 and March 5 is 4 days; if there is another purchase application on March 10, the execution interval between it and March 5 is 5 days. By organizing and analyzing the above time data, a time distribution feature is constructed for each process node identifier, including the execution start time, execution duration, and execution interval period.

[0021] Step S113: Extract the sequence of operation actions corresponding to each process node identifier from the user operation behavior data, and generate the operation association feature for each process node identifier. The operation association feature includes the operation triggering condition, operation execution order and operation result feedback.

[0022] In user operation behavior data, the sequence of operation actions corresponding to each process node identifier needs to be extracted and analyzed to generate operation association features. For process node identifier "CG-SQ-001" (purchase request), the operation trigger condition may be that the inventory is below the safety stock threshold. When the inventory management system detects that the inventory of a certain material is below the set threshold, it triggers the purchasing specialist to perform a purchase request operation. The operation execution sequence is as follows: first, log in to the purchasing management system, enter the purchase request module, fill in the material name, specifications, quantity, expected delivery time, etc., then upload the relevant requirement specification document, and finally click the "Submit" button to complete the application. The operation result feedback includes the system displaying a "Application submitted successfully" prompt, generating a purchase request number, and sending the application to the approver. For process node identifier "CG-XZ-002" (supplier selection), the operation trigger condition is that the purchase request is approved. The operation execution sequence is as follows: view the approved purchase request, filter suppliers that meet the material requirements from the supplier database, view the quotation, qualifications, delivery cycle, etc. of each supplier, compare and evaluate them, select a suitable supplier, and record the selection result. The operation result feedback is that the system records the supplier selection result and generates a supplier selection report. By analyzing the above sequence of operations, operational association features are formed for each process node.

[0023] Step S114: Parse the document transfer records corresponding to each process node identifier in the document flow data, and generate document association features for each process node identifier. The document association features include document type, transfer path and transfer time.

[0024] In the document flow data, each process node identifier corresponds to a different document transfer record. Document association features are generated by parsing these records. Taking the process node identifier "CG-HT-003" (Contract Signing) as an example, the corresponding document type is mainly a purchase contract, including electronic contracts and scanned copies of paper contracts. The transfer path is as follows: after the purchasing department drafts the contract, it is passed to the supplier for confirmation. After the supplier confirms that there are no errors, it is returned to the purchasing department. The purchasing department then passes the contract to the legal department for review. After the legal department approves it, it is passed to the finance department for filing, and finally returned to the purchasing department for archiving. In terms of transfer timeliness, the average time from the purchasing department drafting the contract to the supplier's confirmation is 2 working days, the legal department's review takes an average of 1 working day, and the finance department's filing takes an average of 0.5 working days. For the process node identifier "CG-RK-005" (Warehouse Acceptance), the document type is a warehouse acceptance form. The transfer path is that after the warehouse department accepts the materials, it fills out an acceptance form, passes it to the purchasing department for confirmation, and after the purchasing department confirms it, it is passed to the finance department as a basis for payment. The transfer timeliness is usually completed within 1 working day after acceptance. By analyzing the above document transmission records, the document type, transmission path, and transmission time of each process node are identified, thus forming document association characteristics.

[0025] Step S115: Mine the association of collaborative entities corresponding to each process node identifier in the collaborative interaction data, and generate collaborative association features for each process node identifier. The collaborative association features include collaborative entity type, interaction frequency and interaction content association.

[0026] By mining the collaborative interaction data, the associations of collaborative entities corresponding to each process node identifier are generated to produce collaborative association features. For process node identifier "CG-XZ-002" (supplier selection), the collaborative entity types include purchasing specialists (internal entity) and suppliers (external entity). In terms of interaction frequency, during the supplier selection stage, the average interaction frequency between the purchasing specialist and each potential supplier is 5 times, including telephone communication and email exchanges. The interaction content mainly revolves around material prices, quality standards, delivery cycles, and payment methods. For example, the purchasing specialist inquires about quotations from suppliers, suppliers reply with quotations, and the purchasing specialist communicates regarding questions in the quotations. For process node identifier "CG-FK-006" (financial payment), the collaborative entity types are the purchasing department and the finance department (both internal entities). The interaction frequency is an average of 3 times before each payment, mainly the purchasing department providing payment applications, invoices, acceptance forms, and other documents to the finance department, and the finance department communicating on document issues during the review process. The interaction content involves payment amount, payment time, invoice authenticity, and acceptance status. By mining collaborative interaction data, we can determine the type of collaborative subject, interaction frequency, and interaction content association of each process node, thus forming collaborative association characteristics.

[0027] Step S116: Integrate the time distribution features, operation association features, document association features, and collaboration association features of each process node identifier to form a comprehensive association feature set for each process node identifier.

[0028] The time distribution characteristics, operation association characteristics, document association characteristics, and collaboration association characteristics of each process node are integrated to form a comprehensive association feature set. Taking the process node identifier "CG-SQ-001" (purchase application) as an example, its comprehensive association feature set includes: time distribution characteristics such as the set of execution start times (e.g., 9:15 AM on March 1, 2024, 10:03 AM on March 5, 2024), the set of execution duration periods (25 minutes, 27 minutes, etc.), and the set of execution interval periods (4 days, 5 days, etc.); operation association characteristics such as operation trigger conditions (inventory is below the safety stock threshold), operation execution sequence (login to the system - fill in information - upload document - submit), and operation result feedback (successful submission prompt, generation of application number); document association characteristics such as document type (purchase application form), transmission path (purchasing specialist - approver), and transmission timeliness (immediately transmitted to the approver after submission); and collaboration association characteristics such as the type of collaborating entity (purchasing specialist, approver), interaction frequency (an average of 1 interaction with the approver), and interaction content association (communication of approval opinions). By integrating the features from the different dimensions mentioned above, a comprehensive set of associated features for the process node identifier is formed, which fully reflects the associated information of the node in the business process.

[0029] Step S117: Based on the comprehensive associated feature set of all process node identifiers, calculate the feature similarity between any two process node identifiers. The feature similarity is generated based on the matching degree of each associated feature.

[0030] Step S1171: Extract the time distribution features of any two process node identifiers, calculate the similarity indices of the execution start time difference, the overlap ratio of execution duration, and the consistency of execution interval cycle, and combine the three similarity indices to generate the time dimension evaluation result.

[0031] Select any two process node identifiers, such as "CG-SQ-001" (purchase request) and "CG-XZ-002" (supplier selection), and extract their time distribution characteristics. Calculate the execution start time difference, i.e., the average difference between the execution start time of "CG-SQ-001" and the execution start time of "CG-XZ-002" across multiple executions, for example, an average difference of 2 days. Analyze the overlap of execution durations, assuming the average execution duration of "CG-SQ-001" is 25 minutes and that of "CG-XZ-002" is 60 minutes, there is no overlap, and the overlap ratio is 0. Calculate the similarity of execution interval cycles, if the average execution interval cycle of "CG-SQ-001" is 4.5 days and that of "CG-XZ-002" is 5 days, and obtain the similarity index by calculating the relative difference in their interval cycles. These three similarity indicators are combined according to certain weights, such as the difference in execution start time accounting for 30% of the weight, the overlap ratio of execution duration accounting for 40% of the weight, and the consistency of execution interval period accounting for 30% of the weight. The time dimension evaluation result is generated after comprehensive calculation.

[0032] Step S1172: Extract the operation association features of any two process node identifiers, calculate the similarity index of operation trigger condition similarity, operation execution order fit, and operation result feedback association respectively, and combine the three similarity indexes to generate operation dimension evaluation results.

[0033] Taking "CG-SQ-001" (purchase request) and "CG-RK-005" (warehousing and acceptance) as examples, we extract their operational correlation features. The similarity of operation trigger conditions is as follows: "CG-SQ-001" is triggered by inventory falling below the safety stock threshold, while "CG-RK-005" is triggered by material arrival. The trigger conditions are different, resulting in low similarity. The similarity of operation execution order is also analyzed: "CG-SQ-001" involves logging in, filling out the form, uploading, and submitting, while "CG-RK-005" involves checking materials, inspecting quality, filling out the acceptance form, and submitting. The execution order differs somewhat, resulting in moderate similarity. Finally, the correlation of operation result feedback is low: "CG-SQ-001" results in the generation of a purchase request number, while "CG-RK-005" results in the generation of an acceptance form. These three similarity indicators are then weighted appropriately (e.g., each accounting for one-third) to generate an operational dimension evaluation result.

[0034] Step S1173: Extract the document association features of any two process node identifiers, calculate the similarity indicators of document type matching degree, transmission path overlap ratio and transmission timeliness fit, and combine the three similarity indicators to generate document dimension evaluation results.

[0035] Taking "CG-HT-003" (contract signing) and "CG-FK-006" (financial payment) as examples, document association features are extracted. Document type matching: The document type of "CG-HT-003" is a purchase contract, while the document types of "CG-FK-006" include invoices, payment applications, etc., indicating a low matching degree between contract and invoice document types. Overlapping transmission paths: The transmission path of "CG-HT-003" involves the purchasing department, supplier, legal department, and finance department, while the transmission path of "CG-FK-006" involves only the purchasing department and finance department. The purchasing and finance departments overlap in their transmission paths. The overlap ratio is calculated based on the number of departments involved. For example, "CG-HT-003" involves 4 departments, while "CG-FK-006" involves 2 departments, resulting in an overlap of 2 departments and an overlap ratio of (2 / 4 + 2 / 2) / 2 = 0.75. The delivery timeliness is considered: the overall delivery time for "CG-HT-003" is approximately 3.5 working days, while the delivery time for "CG-FK-006" is approximately 2 working days. The relative difference in delivery time between the two is calculated to obtain the matching index. These three indicators are then combined to generate a document-level evaluation result.

[0036] Step S1174: Extract the collaboration association features of any two process node identifiers, calculate the similarity index of collaboration subject type similarity, interaction frequency association and interaction content matching degree respectively, and combine the three similarity indexes to generate collaboration dimension evaluation results.

[0037] We selected "CG-XZ-002" (supplier selection) and "CG-HT-003" (contract signing) to extract collaboration-related features. For the similarity of collaboration subject types, "CG-XZ-002" involves a purchasing specialist and a supplier, while "CG-HT-003" involves a purchasing specialist, a supplier, and the legal department. Both include purchasing specialists and suppliers, indicating a high degree of similarity in subject types. Regarding the correlation of interaction frequency, "CG-XZ-002" averages 5 interactions with suppliers, while "CG-HT-003" averages 3 interactions. The correlation coefficient between these interaction frequencies was calculated to obtain the correlation index. Finally, regarding the matching degree of interaction content, "CG-XZ-002" focuses on price and qualifications, while "CG-HT-003" focuses on contract terms and breach of contract liabilities. The content is somewhat related but not entirely identical, indicating a moderate matching degree. We then combined these three indicators to generate a collaboration dimension evaluation result.

[0038] Step S1175: Extract data flow records and resource sharing records related to the two process node identifiers from the multimodal process data, calculate the data flow frequency correlation similarity index and the resource sharing depth similarity index, and comprehensively generate data and resource dimension evaluation results.

[0039] For "CG-SQ-001" (purchase requisition) and "CG-XZ-002" (supplier selection), relevant data flow records and resource sharing records are extracted from the multimodal process data. The correlation of data flow frequency is calculated by statistically analyzing the frequency of output data (purchase requisition) from "CG-SQ-001" to "CG-XZ-002" and the frequency of related data feedback from "CG-XZ-002" to "CG-SQ-001," thus determining the degree of correlation between the two data flow frequencies. The depth of resource sharing is analyzed by examining the types of resources shared by the two nodes, such as supplier databases and material information databases, as well as the frequency of use and dependence on these shared resources, generating a resource sharing depth similarity index. The data flow frequency correlation and resource sharing depth similarity index are then combined with certain weights to obtain the data and resource dimension evaluation results.

[0040] Step S1176: Based on the evaluation results of the time dimension, operation dimension, document dimension, collaboration dimension, and data and resource dimension, a comprehensive analysis is performed using preset association rules to obtain the feature similarity between any two process node identifiers. The feature similarity calculation between all any two process node identifiers is completed to form a feature similarity matrix.

[0041] Pre-defined association rules are established, such as assigning weights of 20% to the evaluation results for the time, operation, document, collaboration, and data & resource dimensions, respectively. The evaluation results for each dimension of "CG-SQ-001" and "CG-XZ-002" are then weighted and summed according to these weights to obtain the feature similarity between them. Using the same method, the feature similarity between any two process node identifiers is calculated. These similarity values ​​are then arranged in a matrix, with rows and columns corresponding to each process node identifier, and matrix elements representing the feature similarity between the corresponding two node identifiers, forming a feature similarity matrix.

[0042] Step S118: Based on feature similarity, establish a connection between any two process node identifiers. The connection is determined based on the degree of conformity of feature similarity.

[0043] A feature similarity threshold is set, for example, 0.6. When the feature similarity between any two process node identifiers is greater than or equal to this threshold, a connection is considered to exist between them; when the feature similarity is less than this threshold, a connection is considered not to exist. For example, the feature similarity calculation result for "CG-SQ-001" (purchase request) and "CG-XZ-002" (supplier selection) is 0.75, which is greater than 0.6, therefore a connection is established between them; while the feature similarity between "CG-SQ-001" and "CG-FK-006" (financial payment) is 0.4, which is less than 0.6, so no connection is established. The connection relationships between all process node identifiers are determined in this way.

[0044] Step S119: Calculate the total data flow and resource usage ratio corresponding to each connection relationship, and generate data flow intensity and resource usage association. The data flow intensity is generated based on the total data transmission volume, and the resource usage association is generated based on the usage ratio of shared resources.

[0045] For two process nodes with established connections, such as "CG-SQ-001" and "CG-XZ-002", the total amount of data transferred from "CG-SQ-001" to "CG-XZ-002" within a certain period is statistically analyzed. This includes the number of purchase requisitions and the data size. The total amount of data is then quantified to generate data flow intensity, for example, expressed as "data volume / unit time". Regarding resource usage correlation, the shared resources of these two nodes are analyzed, such as system server resources and database storage space. The proportion of shared resources occupied by each node is calculated. For example, if "CG-SQ-001" occupies 30% of the shared server resources and "CG-XZ-002" occupies 25%, this proportional relationship is used as a quantitative indicator of resource usage correlation.

[0046] Step S1110: Construct a process node association matrix using process node identifiers as matrix rows and columns, and connection relationships, data flow intensity, and resource consumption associations as matrix elements.

[0047] All process node identifiers are used as rows and columns of a matrix, with both row and column indices representing the process node identifiers. For each element (i, j) in the matrix, where i and j are the process node identifiers corresponding to the row and column, if a connection exists between i and j, the element contains a connection identifier (e.g., "1" indicates a connection, "0" indicates no connection), a data flow intensity value, and a resource consumption correlation value. If no connection exists, the connection identifier is "0," and the data flow intensity and resource consumption correlation values ​​are both "0." For example, the element at position (CG-SQ-001, CG-XZ-002) in the matrix is ​​(1, 50 copies / day, 30%:25%), indicating a connection between these two nodes, a data flow intensity of 50 copies per day, and a resource consumption correlation of 30% and 25% respectively. A process node correlation matrix is ​​constructed using this method.

[0048] Step S120: Based on the process node association matrix and the real-time running status of the business process, the process node association network is dynamically evolved to generate a dynamically evolved association network. The dynamically evolved association network includes the dynamic change trajectory of node association strength, link evolution trend and network structure adaptability characteristics.

[0049] In enterprise procurement processes, the dynamic evolution of the process node association network is achieved based on an established process node association matrix and real-time operational status. For example, real-time operational status might show that the current procurement request stage (CG-SQ-001) is lagging behind, or that the supplier selection stage (CG-XZ-002) is experiencing excessive resource consumption. These real-time statuses affect the association relationships and strength between nodes, thus causing the association network to evolve. By analyzing and processing the aforementioned real-time data, a dynamically evolving association network that reflects the dynamic changes in the network is generated.

[0050] Step S121: Construct the initial process node association network by using the process node identifiers in the process node association matrix as nodes of the initial process node association network, the connection relationship as the edge of the initial process node association network, and the data flow intensity and resource consumption association as edge attributes of the initial process node association network.

[0051] Each process node identifier in the process node association matrix is ​​used as a node in the initial process node association network. Based on the connections in the matrix, edges are established between corresponding nodes. For example, if there is a connection between (CG-SQ-001, CG-XZ-002) in the matrix, an edge is added between nodes CG-SQ-001 and CG-XZ-002 in the initial network. The edge attributes are set to data flow intensity and resource consumption association, such as the edge attributes of 50 data flows / day and 30%:25% resource consumption association. The initial process node association network is constructed in this way, and this network can initially reflect the association between each process node.

[0052] Step S122: Collect real-time running status data of the business process. The real-time running status data includes the real-time execution progress of each process node, the real-time resource occupancy, and the real-time data flow rate.

[0053] The enterprise's business process management system collects real-time operational status data for each process node. For the procurement requisition stage (CG-SQ-001), real-time execution progress can be the ratio of the number of currently submitted procurement requisitions to the planned number; real-time resource utilization includes system memory and CPU usage; and real-time data flow rate is the number of procurement requisitions submitted per unit time. For the supplier selection stage (CG-XZ-002), real-time execution progress is the ratio of the number of suppliers that have completed screening to the total number of suppliers to be screened; real-time resource utilization includes database query resource utilization and network bandwidth utilization; and real-time data flow rate is the amount of supplier information data processed per unit time. Similarly, real-time execution progress, resource utilization, and data flow rate are collected for other process nodes such as contract signing (CG-HT-003), order placement (CG-DD-004), warehousing and acceptance (CG-RK-005), and financial payment (CG-FK-006) to form complete real-time operational status data.

[0054] Step S123: Extract the difference between the real-time execution progress and the historical execution progress of each process node in the real-time running status data, and generate node progress deviation features. The node progress deviation features are generated based on the degree of deviation between the real-time progress and the historical average progress.

[0055] For each process node, the real-time execution progress is extracted from the real-time running status data, and the historical average execution progress data for that node is obtained. For example, the current real-time execution progress of the procurement requisition process (CG-SQ-001) is 60%, while the historical average execution progress over the same period is typically 75%. Therefore, the difference is 60% - 75% = -15%. This difference value can be used to represent the node progress deviation characteristic; a negative difference value indicates a delay, while a positive value indicates a lead. The difference values ​​of each process node are integrated to generate the node progress deviation characteristic, which reflects the deviation of each node's execution progress from the historical average level.

[0056] Step S124: Analyze the degree of conformity between the real-time resource occupancy of each process node and the preset resource configuration standard, and generate node resource occupancy deviation characteristics. The node resource occupancy deviation characteristics are generated based on the difference between the real-time occupancy and the preset configuration.

[0057] Resource configuration standards are preset for each process node. For example, in the procurement application stage (CG-SQ-001), the preset CPU utilization standard is 20%, and the real-time utilization is 25%, so the difference is 25% - 20% = 5%. In the supplier selection stage (CG-XZ-002), the preset database query resource utilization standard is 30 units / second, and the real-time utilization is 35 units / second, with a difference of 5 units / second. The node resource utilization deviation characteristic is generated by the difference between these real-time utilization values ​​and the preset configuration values. A positive difference indicates that the resource utilization exceeds the standard, and a negative difference indicates that the standard is not met. The resource utilization deviation values ​​of each node are integrated to form the node resource utilization deviation characteristic.

[0058] Step S125: Calculate the ratio of the real-time data flow rate to the historical average flow rate of each process node, and generate node data flow deviation characteristics. The node data flow deviation characteristics are generated based on the ratio between the real-time rate and the historical rate.

[0059] Obtain the real-time data flow rate and historical average flow rate for each process node, and calculate their ratio. For example, in the warehousing and acceptance stage (CG-RK-005), the real-time data flow rate is 10 transactions / hour, and the historical average flow rate is 15 transactions / hour, with a ratio of 10 / 15 ≈ 0.67. In the financial payment stage (CG-FK-006), the real-time data flow rate is 8 transactions / hour, and the historical average flow rate is 6 transactions / hour, with a ratio of 8 / 6 ≈ 1.33. The node data flow deviation characteristic is represented by this ratio; a ratio less than 1 indicates a flow rate lower than the historical average, and a ratio greater than 1 indicates a flow rate higher than the historical average. Integrate the ratios of each node to generate the node data flow deviation characteristic.

[0060] Step S126: Integrate node progress deviation characteristics, node resource occupancy deviation characteristics, and node data flow deviation characteristics to form a node operation status deviation set; based on the node operation status deviation set, process it through a preset driving analysis model to generate a network evolution driving factor, which is a parameter used to comprehensively reflect the degree of influence of node state changes on network correlation strength.

[0061] By integrating node progress deviation characteristics, node resource utilization deviation characteristics, and node data flow deviation characteristics, a node operational status deviation set is formed. This set contains deviation information for each process node in terms of progress, resource utilization, and data flow. The pre-defined driving analysis model can be a multi-factor comprehensive evaluation model. This model takes the various deviation values ​​in the node operational status deviation set as input and processes them through algorithms (such as weighted summation, neural network calculation, etc.) to generate network evolution driving factors. For example, the model assigns higher weights to nodes with delayed progress or excessive resource utilization, calculating a comprehensive driving factor value. The larger the driving factor value, the greater the impact of node state changes on network correlation strength.

[0062] Step S127: Input the network evolution driving factor into the initial process node association network, adjust the attribute values ​​of the edges in the initial process node association network, update the node association strength in the initial process node association network, and generate the process node association network after the first evolution.

[0063] Step S1271: Extract the specific values ​​of node progress deviation characteristics, node resource occupancy deviation characteristics, and node data flow deviation characteristics from the network evolution driving factors.

[0064] The specific values ​​of node progress deviation, node resource utilization deviation, and node data flow deviation are extracted from the network evolution driving factors. For example, for the procurement requisition stage (CG-SQ-001), the node progress deviation value is -15%, the node resource utilization deviation value is 5%, and the node data flow deviation value is 0.8 (assuming that the ratio of its real-time data flow rate to the historical average rate is 0.8). These specific values ​​will be used for subsequent edge attribute adjustments.

[0065] Step S1272: Analyze the impact of the node progress deviation characteristic value on the connection relationship of the corresponding process node, and determine the edge attribute adjustment ratio of the initial process node association network corresponding to the node progress deviation characteristic. The adjustment ratio is generated based on the magnitude of the deviation characteristic value.

[0066] For the characteristic values ​​of node schedule deviations, analyze their impact on the connection relationships between process nodes. For example, the schedule deviation of the procurement requisition stage (CG-SQ-001) is -15% (lagging), which may affect its connection relationship with subsequent nodes (such as CG-XZ-002), and the data flow intensity may need to be reduced. Set adjustment rules, such as a 5% reduction in the adjustment ratio of edge attributes for every 10% delay in schedule deviation. Then, the adjustment ratio corresponding to a deviation of -15% is -7.5% (i.e., a reduction of 7.5%).

[0067] Step S1273: Determine the adjustment range of the corresponding resource usage association attribute based on the node resource usage deviation characteristic value. The adjustment range is generated based on the ratio between the deviation characteristic value and the preset deviation threshold.

[0068] A preset resource utilization deviation threshold is set, such as a 10% deviation threshold for CPU utilization. For the procurement requisition stage (CG-SQ-001), the resource utilization deviation characteristic value is 5%, which is 50% of the 10% threshold. The adjustment range rule is set to half of this ratio, i.e., an adjustment range of 25%. If the resource utilization deviation is positive (exceeding the standard), the adjustment range for the associated resource utilization attribute is reduced by this range; if it is negative, the adjustment range is increased. Here, a 5% deviation corresponds to an adjustment range of 25%.

[0069] Step S1274: Based on the node data flow deviation characteristic value, determine the adjustment coefficient of the corresponding data flow intensity attribute. The adjustment coefficient is generated based on the sign and magnitude of the deviation characteristic value.

[0070] The node data flow deviation characteristic value is the ratio of the real-time rate to the historical rate. If the ratio is 0.8 (e.g., CG-SQ-001), it indicates that the data flow rate is lower than the historical average. An adjustment coefficient rule is set, with the adjustment coefficient equal to this ratio. Therefore, the adjustment coefficient for the data flow intensity attribute is 0.8, meaning the data flow intensity will be adjusted by multiplying by 0.8.

[0071] Step S1275: Integrate the adjustment ratio, adjustment magnitude, and adjustment coefficient into a set of edge attribute adjustment parameters. The set of edge attribute adjustment parameters includes the adjustment parameters of each attribute of the edge of the network associated with each initial process node.

[0072] The adjustment ratios, magnitudes, and coefficients for each node are then integrated along the edges. For example, for the edge between CG-SQ-001 and CG-XZ-002, the adjustment ratio is -7.5% (based on schedule deviation), the resource usage-related adjustment magnitude is a reduction of 25% (based on resource deviation), and the data flow intensity adjustment coefficient is 0.8 (based on data flow deviation). These parameters are then integrated into an edge attribute adjustment parameter set, where each edge has corresponding adjustment parameters.

[0073] Step S1276: Traverse all edges in the initial process node association network and locate the process node identifier combination corresponding to each edge in the initial process node association network.

[0074] Examine each edge in the initial process node association network in turn, and determine the combination of the two process node identifiers connected by each edge. For example, one edge connects CG-SQ-001 and CG-XZ-002, and another edge connects CG-XZ-002 and CG-HT-003, etc., and record the above node identifier combinations.

[0075] Step S1277: Match the corresponding adjustment parameters from the set of edge attribute adjustment parameters based on the combination of process node identifiers.

[0076] For each combination of process node identifiers corresponding to an edge, the corresponding adjustment parameters are searched in the set of edge attribute adjustment parameters. For example, for the node identifier combination CG-SQ-001 and CG-XZ-002, the following parameters are matched from the set: adjustment ratio -7.5%, resource consumption related adjustment range reduced by 25%, and data flow intensity adjustment coefficient 0.8.

[0077] Step S1278: Modify the data flow intensity attribute value and resource consumption association attribute value of the edges of the network associated with each initial process node according to the adjustment parameters.

[0078] Taking the edge between CG-SQ-001 and CG-XZ-002 as an example, the original data flow intensity was 50 units / day, with an adjustment coefficient of 0.8. The modified data flow intensity is 50 × 0.8 = 40 units / day. The original resource usage ratio was 30%:25%, and the adjustment range is a reduction of 25%. Therefore, the adjusted ratio is (30% × (1-25%)): (25% × (1-25%)) = 22.5%:18.75%. Modify the attribute values ​​of each edge in the same way.

[0079] Step S1279: Based on the adjusted edge attribute values, recalculate the node association strength between the two process node identifiers connected by the edge of each initial process node association network. The node association strength is generated based on the adjusted attribute values.

[0080] The node association strength can be calculated by comprehensively considering the adjusted edge attribute values ​​(data flow intensity and resource consumption association). For example, let's set data flow intensity to account for 60% of the weight and resource consumption association similarity to account for 40% of the weight. For the adjusted edges CG-SQ-001 and CG-XZ-002, the data flow intensity is 40 units / day (assuming a full score of 50 units / day, the score is 40 / 50=0.8), and the resource consumption association similarity is calculated by the closeness of the two ratios (22.5% and 18.75% similarity are set to 0.7). Then, the node association strength is 0.8×60%+0.7×40%=0.76.

[0081] Step S12710: Update all edge attribute values ​​and node association strengths in the initial process node association network to form the process node association network after the first evolution.

[0082] The adjusted attribute values ​​of all edges and the recalculated node association strength are updated in the initial process node association network, replacing the original attribute values ​​and association strengths, thus forming the process node association network after the first evolution.

[0083] Step S128: Based on the preset evolution time window, repeatedly collect real-time running status data, generate network evolution driving factors, and adjust the network attributes associated with process nodes to form a process node associated network sequence after multiple rounds of evolution.

[0084] The preset evolution time window is 1 hour, meaning that a network evolution operation is performed once every hour. Within each time window, following steps S122 to S127, real-time operating status data is collected, network evolution driving factors are generated, and the network attributes associated with process nodes are adjusted to obtain the evolved network. After multiple time windows (e.g., once per hour within an 8-hour working period, for a total of 8 times), 8 evolved process node association networks are formed. These networks are arranged in chronological order, forming a process node association network sequence.

[0085] Step S129: Extract the change data of the association strength of each node in the process node association network sequence, and generate the dynamic change trajectory of the node association strength. The dynamic change trajectory of the node association strength is a continuous change curve of the association strength under each time window.

[0086] From the process node association network sequence, the node association strength data for each edge (i.e., node pair) under different time windows is extracted. For example, the node association strength between CG-SQ-001 and CG-XZ-002 is 0.76 in the first hour, 0.72 in the second hour, and 0.68 in the third hour, etc. Arranging the above data in chronological order, the dynamic change trajectory of node association strength is plotted, which can intuitively reflect the change of node association strength over time.

[0087] Step S1210: Analyze the changes in the link structure of the process node association network sequence, generate the link evolution trend, and calculate the degree of fit between the process node association network structure and business process requirements at each evolution stage, generate network structure adaptability characteristics, and integrate the dynamic change trajectory of node association strength, link evolution trend and network structure adaptability characteristics to form a dynamic evolution association network.

[0088] This analysis examines changes in the link structure within the process node association network sequence, such as the addition and deletion of links, and changes in node connection order. It summarizes link evolution trends, for example, the association strength of some links gradually increases while that of others gradually weakens. The degree of fit between the network structure and business process requirements at each evolution stage is calculated. Business process requirements include the execution order of each stage, data flow requirements, and resource allocation standards. By comparing the network structure with these requirements, the degree of fit is quantified (e.g., a fit of 0.8 indicates 80% compliance), generating network structure adaptability characteristics. The dynamic trajectory of node association strength changes, link evolution trends, and network structure adaptability characteristics are integrated to form a dynamically evolving association network. This network comprehensively reflects the dynamic changes in node associations within the business process.

[0089] Step S130: Perform bottleneck gene localization processing on the dynamic evolution association network, and extract bottleneck gene features in the process node association. The bottleneck gene features include node association obstruction features, resource allocation imbalance features, and link evolution stagnation features.

[0090] In dynamically evolving interconnected networks, there are bottleneck factors that affect the smooth operation of business processes. These factors are identified through bottleneck gene localization. For example, it might be found that the node correlation strength between purchase requests and supplier selection is continuously decreasing, leading to data flow obstruction; or that the resource utilization of certain nodes is severely unbalanced, affecting the overall process efficiency; or that the evolution rate of certain links is too slow, resulting in lag. By locating and analyzing the above problems, the corresponding bottleneck gene features are extracted.

[0091] Step S131: Extract the dynamic change trajectory of node association strength in the dynamic evolution association network, and filter the node pairs whose association strength continues to decrease. The node pair is a combination of two process node identifiers with a direct connection relationship.

[0092] Traverse all node pairs (combinations of two process node identifiers with direct connections) in the dynamically evolving association network and examine the dynamic trajectory of their node association strength changes. Filter out node pairs whose association strength continuously decreases over multiple consecutive time windows. For example, the node association strength of CG-XZ-002 (supplier selection) and CG-HT-003 (contract signing) decreased from 0.8 to 0.7, and then to 0.6 over the past three time windows, showing a continuous downward trend; therefore, this node pair is filtered out.

[0093] Step S132: Analyze the link evolution trend corresponding to the node pairs with continuously decreasing association strength, and determine the stagnation period of link evolution. The stagnation period is a continuous time interval in which the rate of decrease in association strength exceeds a preset threshold.

[0094] For node pairs with continuously decreasing association strength, such as CG-XZ-002 and CG-HT-003, analyze their corresponding link evolution trends. Calculate the rate of decrease in association strength within each time window, setting a preset threshold of 0.1 decrease per hour. If, within a continuous time interval, the rate of decrease in association strength for this node pair exceeds 0.1, for example, from hour 2 to hour 4, the association strength decreases from 0.8 to 0.6, with an average decrease of 0.1 per hour, and the rate of decrease in each hour exceeds 0.1, then the time interval from hour 2 to hour 4 is defined as the stagnation period of link evolution.

[0095] Step S133: Based on the node progress deviation characteristics and node data flow deviation characteristics during the stagnation period, generate node-related stagnation characteristics. The node-related stagnation characteristics include the stagnation start time, stagnation duration, and stagnation impact range.

[0096] Step S1331: Extract the start and end times of the stagnation period, and determine the start time as the stagnation start time.

[0097] For the aforementioned stagnation period (from the 2nd to the 4th hour), the starting time is the 2nd hour, and this time point is determined as the starting moment of the blockage.

[0098] Step S1332: Calculate the time difference between the end time point and the start time point, and determine the duration of the blockage.

[0099] The end time is the 4th hour, the start time is the 2nd hour, and the time difference is 4-2=2 hours. We determine 2 hours as the duration of the blockage.

[0100] Step S1333: Extract the process node identifiers corresponding to the node pairs whose association strength continuously decreases during the stagnation period, and record the two process node identifiers of each node pair.

[0101] During the stagnation period, the node pairs with continuously decreasing correlation strength are CG-XZ-002 and CG-HT-003. Record the identifiers of these two process nodes.

[0102] Step S1334: Analyze the number of connected nodes and the length of the connection links for each process node in the dynamic evolution association network to determine the association radiation range of each node.

[0103] Analyzing the connection nodes of CG-XZ-002 in the dynamic evolutionary association network, assuming it connects to CG-SQ-001, CG-HT-003, the supplier database node, etc., with a total of 3 connected nodes and an average link length of 2 (i.e., an average of 2 nodes between each other), its association radiation range includes these connected nodes and their associated links. Similarly, analyzing the number of connected nodes and link length of CG-HT-003, its association radiation range is determined.

[0104] Step S1335: Based on the associated radiation range of each process node identifier in the node pair, calculate the intersection and union regions of the two radiation ranges.

[0105] The associated radiation ranges of CG-XZ-002 and CG-HT-003 are superimposed to calculate their intersection region (the part covered by both ranges) and union region (the whole after merging the two ranges). For example, the intersection region may include some common document processing nodes, while the union region includes the connection nodes and links of each node.

[0106] Step S1336: Extract all process node identifiers within the intersection area and determine the business execution stage corresponding to the process node identifier.

[0107] Extract all process node identifiers from the intersection area, such as document review nodes and information entry nodes, and determine the business execution links corresponding to these node identifiers, such as contract document review links and supplier information entry links.

[0108] Step S1337: Analyze the logical correlation between the business execution links and nodes and the corresponding business execution links, and filter the process node identifiers whose logical correlation meets the preset standards.

[0109] Analyze the logical correlation between the business execution steps (such as contract document review) within the intersection area and the corresponding business execution steps (supplier selection and contract signing) of the node pair (CG-XZ-002 and CG-HT-003). For example, contract document review is a prerequisite step for contract signing, and the logical correlation is high, meeting the preset criteria (such as correlation greater than 0.7), then the process node identifier corresponding to this business execution step is selected.

[0110] Step S1338: Integrate the business execution links corresponding to the filtered process node identifiers, and determine the coverage of the business execution links.

[0111] The selected process nodes are integrated to identify the corresponding business execution steps, such as contract document review and supplier information entry. The scope of these steps in the entire business process is determined, including pre-contract signing preparations and information verification.

[0112] Step S1339: Determine the depth of the impact of the blockage on the business execution process within the coverage area by combining the decrease in data flow intensity during the duration of the blockage.

[0113] During the two-hour disruption, the data transfer intensity of nodes CG-XZ-002 and CG-HT-003 decreased from 30 copies / day to 20 copies / day, a reduction of 10 copies / day. Based on the magnitude of the decrease, the depth of impact on business execution processes within the coverage area was assessed. For example, if the depth of impact was moderate, it would result in a 30% reduction in the execution efficiency of relevant processes.

[0114] Step S13310: Generate node-related blocking features based on the blocking start time, blocking duration, coverage area, and influence depth.

[0115] By integrating the start time of the blockage (2nd hour), duration of the blockage (2 hours), scope of coverage (preparatory work and information verification work before contract signing), and depth of impact (moderate, efficiency reduced by 30%), node-related blockage characteristics are generated.

[0116] Step S134: Extract resource occupancy correlation data of each process node in the dynamic evolution correlation network, analyze the changes in the allocation ratio of the same resource type among different process nodes, and filter out resource types whose allocation ratio fluctuations exceed the preset range.

[0117] Resource usage correlation data for each process node is extracted from the dynamic evolutionary correlation network, such as the allocation ratio of CPU resources, memory resources, and database connection resources among different process nodes. The distribution ratio of the same resource type (e.g., CPU resources) across each process node changes over time, with a preset range of ±15%. If the allocation ratio of CPU resources between CG-XZ-002 (supplier selection) and CG-RK-005 (warehousing and acceptance) changes from 40%:30% to 60%:10%, a fluctuation of 20%, exceeding the preset range of ±15%, then the aforementioned resource type of CPU resources is selected.

[0118] Step S135: Based on the node resource occupancy deviation characteristics corresponding to the resource type, identify the first process node cluster whose real-time resource occupancy is consistently higher than the preset configuration amount, and the second process node cluster whose real-time resource occupancy is consistently lower than the preset configuration amount.

[0119] For the selected CPU resource types, examine the node resource usage deviation characteristics of each process node. Identify process nodes whose real-time resource usage consistently exceeds the preset configuration within multiple time windows, such as CG-XZ-002 (supplier selection) and CG-HT-003 (contract signing), and group them into the first process node cluster. Simultaneously, identify process nodes whose real-time resource usage consistently falls below the preset configuration, such as CG-RK-005 (warehousing and acceptance) and CG-FK-006 (financial payment), and group them into the second process node cluster.

[0120] Step S136: Based on the resource occupancy of the first process node cluster and the second process node cluster, calculate the resource allocation difference between them; combine the connection relationship of the associated node pairs connected to the first process node cluster and the second process node cluster and the data flow intensity, evaluate the impact range of the resource allocation difference, and generate resource allocation imbalance features. The resource allocation imbalance features include the imbalanced resource type, the identifiers of the first process node cluster and the second process node cluster, the resource allocation difference value, and the impact range evaluation result.

[0121] Calculate the total CPU resource usage of the first process node cluster (CG-XZ-002, CG-HT-003) and the total CPU resource usage of the second process node cluster (CG-RK-005, CG-FK-006). The difference between the two is the resource allocation imbalance value. For example, if the total usage of the first cluster is 70% and the total usage of the second cluster is 20%, the difference is 50%. Analyze the connection relationships of related node pairs connected to the first and second clusters (such as the connection between CG-XZ-002 and CG-SQ-001) and the data flow intensity to assess the impact of the resource allocation imbalance on the data flow and business execution of these nodes, such as affecting the processing speed of purchase requests and the efficiency of warehousing and acceptance. Integrate the imbalanced resource type (CPU resources), the identifiers of the first and second clusters, the resource allocation imbalance value (50%), and the impact assessment results (affecting the processing speed of purchase requests and the efficiency of warehousing and acceptance) to generate a resource allocation imbalance feature.

[0122] Step S137: Extract the link evolution trend data in the dynamic evolution association network, and filter the links whose evolution rate is continuously lower than the preset evolution rate. The links are continuous association paths connecting multiple process nodes.

[0123] Evolutionary trend data for each link is extracted from the dynamic evolutionary network. A link refers to a continuous path connecting multiple process nodes, such as CG-SQ-001→CG-XZ-002→CG-HT-003→CG-DD-004 (purchase request→supplier selection→contract signing→order placement). The preset evolution rate is 0.05 per hour (the increase in association strength). Links whose evolution rate is consistently lower than this preset value are filtered out. For example, the evolution rates of the above links in the past 3 hours were 0.03, 0.02, and 0.01, all lower than 0.05, and therefore were filtered out.

[0124] Step S138: Analyze the dynamic change trajectory of the node association strength and the network structure adaptability characteristics corresponding to the link, and identify the key nodes of the link evolution bottleneck. The key nodes are the core process nodes in the link that cause the evolution rate to decrease.

[0125] For the selected links (CG-SQ-001→CG-XZ-002→CG-HT-003→CG-DD-004) whose evolution rate consistently falls below the preset value, the dynamic trajectory of node association strength and network structure adaptability characteristics of each node in the link were analyzed. It was found that the node association strength of CG-HT-003 (contract signed) decreased significantly, and its network structure adaptability characteristic value was low (e.g., 0.5, below the average level of 0.7). This indicates that this node is the core node causing the decrease in evolution rate in the link; therefore, CG-HT-003 was identified as the key node causing the link evolution stagnation.

[0126] Step S139: Based on the progress deviation characteristics, resource usage deviation characteristics, and data flow deviation characteristics of key nodes, generate link evolution stagnation characteristics. The link evolution stagnation characteristics include the stagnation node identifier, the stagnation start time period, and the stagnation propagation path.

[0127] Extract the schedule deviation characteristics (e.g., 20% lag), resource usage deviation characteristics (CPU usage exceeding 15%), and data flow deviation characteristics (rate ratio 0.6) of the key node CG-HT-003. Based on these characteristics, determine the initiation period of the bottleneck, for example, a significant bottleneck starting from the 3rd hour; the bottleneck propagation path is from CG-HT-003 backwards to CG-DD-004 (order placement), and forwards to CG-XZ-002 (supplier selection). Integrate the bottleneck node identifier (CG-HT-003), the bottleneck initiation period (3rd hour), and the bottleneck propagation path (CG-XZ-002←CG-HT-003→CG-DD-004) to generate the link evolution bottleneck characteristics.

[0128] Step S1310: Integrate the node association blockage characteristics, resource allocation imbalance characteristics, and link evolution stagnation characteristics to form bottleneck gene characteristics in process node association.

[0129] The node association obstruction feature generated in step S133, the resource allocation imbalance feature generated in step S136, and the link evolution bottleneck feature generated in step S139 are integrated together to form the bottleneck gene feature in the process node association. This bottleneck gene feature comprehensively reflects the bottleneck problems such as obstruction, resource imbalance, and bottleneck in the business process.

[0130] Step S140: Construct a simulation adaptation scenario, input the bottleneck gene features into the simulation adaptation scenario for dynamic deduction and optimization, and generate adaptation optimization parameters. The adaptation optimization parameters include node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters.

[0131] A simulation adaptation scenario is constructed to mimic the operation of an enterprise's procurement business process. The extracted bottleneck characteristics are input into this scenario, and different optimization schemes are dynamically deduced to find adaptation optimization parameters that can alleviate or solve the bottleneck problem. For example, for node association obstruction characteristics, different node association strength adjustment schemes are deduced; for resource allocation imbalance characteristics, different resource reallocation schemes are deduced; for link evolution stagnation characteristics, different link evolution guidance schemes are deduced, ultimately generating corresponding adaptation optimization parameters.

[0132] Step S141: Based on the core execution objectives and preset operating standards of the business process, construct a basic simulation scenario. The basic simulation scenario includes basic configuration of process nodes, basic resource allocation scheme and basic link connection mode.

[0133] The core execution objective of the business process is to improve procurement efficiency and reduce procurement costs while ensuring procurement quality. Preset operational standards include the average execution time, resource consumption limits, and data flow rate limits for each process node. Based on these objectives and standards, a basic simulation scenario is constructed. Basic configurations for process nodes include the name, functional description, and input / output data types of each node; for example, the purchase requisition node is configured to receive inventory information input and output a purchase requisition form. The basic resource allocation scheme sets the initial resource allocation for each node, such as CPU resource allocation ratio and memory size. The basic link connection mode defines the connection relationships and data flow between nodes; for example, the purchase requisition node connects to the supplier selection node, and the supplier selection node connects to the contract signing node.

[0134] Step S142: Extract the node association blockage features from the bottleneck gene features, determine the blockage node pairs and the blockage influence range, import the blockage node pairs, the blockage influence range and the corresponding blockage feature parameters into the basic simulation scenario, and generate a simulation sub-scenario containing the blockage problem.

[0135] Node-related blocking features were extracted from bottleneck gene characteristics, identifying the blocking node pair as CG-XZ-002 and CG-HT-003. The blocking impact ranged from pre-contract signing preparations to information verification. Blocking feature parameters included the blocking start time (2nd hour), blocking duration (2 hours), and impact depth (efficiency reduction of 30%). This information was imported into a basic simulation scenario to simulate the blocking relationship between the node pair, generating a simulation sub-scenario that included the blocking problem.

[0136] Step S143: Extract the resource allocation imbalance features from the bottleneck gene features, determine the imbalanced resource type, imbalanced node cluster identifier and imbalance impact degree, import the corresponding information into the basic simulation scenario, and generate a simulation sub-scenario containing the resource imbalance problem.

[0137] Resource allocation imbalance characteristics were extracted, and the imbalanced resource type was determined to be CPU resources. The first process node clusters were identified as CG-XZ-002 and CG-HT-003, and the second process node clusters were identified as CG-RK-005 and CG-FK-006. The degree of imbalance impacted the processing speed of procurement requests and the efficiency of warehousing and acceptance. The above information was imported into the basic simulation scenario to simulate the imbalanced allocation of CPU resources among the above node clusters, generating a simulation sub-scenario that included the resource imbalance problem.

[0138] Step S144: Extract the link evolution stuttering features from the bottleneck gene features, determine the stuttering node identifier, stuttering start time period and stuttering propagation path, import the corresponding information into the basic simulation scenario, and generate a simulation sub-scenario containing the link stuttering problem.

[0139] Extract the link evolution lag features, identify the lag node as CG-HT-003, determine the lag start time as the 3rd hour, and determine the lag propagation path as CG-XZ-002←CG-HT-003→CG-DD-004. Import the above information into the basic simulation scenario to simulate the lag and propagation of this node, generating a simulation sub-scenario that includes the link lag problem.

[0140] Step S145: Integrate the simulation sub-scenarios containing bottleneck problems, simulation sub-scenarios containing resource imbalance problems, and simulation sub-scenarios containing link blockage problems, and overlay the impact and correlation of each problem to generate a complete simulation adaptation scenario. The complete simulation adaptation scenario includes all bottleneck problems in the business process and the interaction relationships between problems.

[0141] The three simulation sub-scenarios mentioned above are integrated to analyze the interactions between the problems. For example, resource imbalance may lead to node association blockage, which in turn may exacerbate link evolution bottlenecks. During the integration process, these mutual influences are superimposed onto the scenario, enabling the scenario to realistically reflect all bottleneck problems and their interactions in the business process, generating a complete simulation adaptation scenario.

[0142] Step S146: Construct multiple sets of node association adjustment schemes. Each set of node association adjustment schemes includes the node association strength adjustment ratio, association method optimization suggestions, and strategies for adjusting the connection relationship of blocking node pairs.

[0143] Multiple node association adjustment schemes were constructed, such as: Scheme 1: The node association strength adjustment ratio is increased by 20% between CG-XZ-002 and CG-HT-003, the association method optimization suggestion is to adopt parallel data transmission, and the strategy for blocking node connection relationship adjustment is to add a backup connection link; Scheme 2: The node association strength adjustment ratio is increased by 15%, the association method optimization suggestion is to compress the data transmission packet size, and the strategy for blocking node connection relationship adjustment is to optimize the bandwidth allocation of the existing connection link; Scheme 3: The node association strength adjustment ratio is increased by 25%, the association method optimization suggestion is to adopt a data caching mechanism, and the strategy for blocking node connection relationship adjustment is to change the connection protocol, etc. A total of 5 different schemes were constructed.

[0144] Step S147: Input each set of node association adjustment schemes into the complete simulation adaptation scenario, run the simulation process, record the change data of node association hindrance characteristics during the simulation, and select a set of schemes from multiple sets of node association adjustment schemes according to the preset screening rules. The node association strength adjustment ratio, association mode adjustment parameters and hindrance node pair connection relationship adjustment instructions contained in the scheme are determined as node association adjustment parameters.

[0145] Step S1471: Import the first set of node association adjustment schemes into the complete simulation adaptation scenario, set the simulation run time step and total duration. The time step is set based on the execution cycle of the business process, and the total duration covers a complete business process execution cycle.

[0146] The first option is imported into the complete simulation adaptation scenario. The execution cycle of the business process is 8 hours, the time step is set to 1 hour, and the total duration is 8 hours to cover a complete execution cycle.

[0147] Step S1472: Start the simulation process and record the changes in various parameters of the node-related hindrance characteristics during the simulation process according to the set time step, including the offset of the hindrance start time, the reduction of the hindrance duration, and the reduction ratio of the hindrance influence range.

[0148] Initiate the simulation process and record the changes in the node-related bottleneck characteristics at each time step (1 hour). For example, if the bottleneck start time was originally the 2nd hour, after implementing Scheme 1, the bottleneck start time shifts to the 3rd hour by an offset of +1 hour; the bottleneck duration is reduced from 2 hours to 1.5 hours by a reduction of 0.5 hours; and the scope of the bottleneck's impact is reduced from covering 5 business execution stages to 3 stages, a reduction of (5-3) / 5 = 40%.

[0149] Step S1473: After the simulation process is completed, evaluate the mitigation effect of the blockage start time, blockage duration and blockage range respectively, and calculate the blockage feature mitigation rate based on the comprehensive mitigation effect of each.

[0150] The evaluation assesses the mitigation effects of the offset at the onset of the blockage (a larger offset is better), the reduction in the duration of the blockage (a larger reduction is better), and the reduction in the area of ​​impact of the blockage (a larger reduction is better). Each factor is weighted at 1 / 3, and the overall blockage feature mitigation rate is calculated. For example, an offset of +1 hour corresponds to a mitigation score of 0.8, a reduction of 0.5 hours corresponds to a score of 0.7, and a reduction of 40% corresponds to a score of 0.6. The overall blockage feature mitigation rate is (0.8 + 0.7 + 0.6) / 3 = 0.7.

[0151] Step S1474: Following the same simulation settings and recording logic, import the remaining node association adjustment schemes into the complete simulation adaptation scenario in sequence, run the simulation process and record the data on the change of the stall characteristics, and calculate the stall characteristic mitigation rate of each scheme.

[0152] Following the same settings and logic as steps S1471 to S1473, import Scheme 2, Scheme 3, Scheme 4, and Scheme 5 respectively, run the simulation process, record the change data of various blocking characteristics, and calculate the relief rate of each blocking characteristic, assuming they are 0.65, 0.75, 0.6, and 0.72 respectively.

[0153] Step S1475: Extract the blockage relief rate data of all group schemes, sort them in descending order of relief rate value, and select the top K schemes as candidate optimization schemes.

[0154] The mitigation rates of the blocking characteristics for all schemes are as follows: Scheme 3 0.75, Scheme 5 0.72, Scheme 1 0.7, Scheme 2 0.65, and Scheme 4 0.6. Schemes 3, 5, and 1 are selected as candidate optimization schemes, sorted from highest to lowest and with K=3.

[0155] Step S1476: Rerun the simulation process of the candidate optimization scheme, increase the number of simulation runs, and obtain the average value of the hindrance feature mitigation rate after multiple rounds of simulation.

[0156] The simulation process was rerun 5 times for each of Schemes 3, 5, and 1, and the mitigation rate of the blocking characteristics was recorded each time. For example, the mitigation rates of Scheme 3 in the 5 simulations were 0.75, 0.76, 0.74, 0.75, and 0.75, respectively, with an average of (0.75+0.76+0.74+0.75+0.75) / 5=0.75; the average value of Scheme 5 was 0.71; and the average value of Scheme 1 was 0.69.

[0157] Step S1477: Calculate the standard deviation of the relief rate of the candidate optimization scheme in multiple rounds of simulation, wherein the standard deviation of the relief rate is generated based on the degree of deviation of the relief rate from the average value in multiple rounds.

[0158] The standard deviation of the remission rate for scheme 3 was calculated. The deviations of each remission rate from the average of 0.75 were 0, 0.01, -0.01, 0, and 0, respectively. The sum of squares was 0.0001 + 0.0001 = 0.0002, and the standard deviation was √(0.0002 / 5) ≈ 0.0063. The standard deviation of scheme 5 was calculated to be 0.012. The standard deviation of scheme 1 was 0.015.

[0159] Step S1478: Based on the average value and standard deviation of the hindrance feature mitigation rate after multiple rounds of simulation, a set of schemes is determined from the candidate optimization schemes according to the preset final selection rules, and the node association strength adjustment ratio, association mode optimization suggestions and hindrance node pair connection relationship adjustment strategies are extracted from the schemes to form node association adjustment parameters.

[0160] The default final selection rule is: prioritize schemes with high average values ​​and low standard deviations. Scheme 3 has the highest average value (0.75) and the lowest standard deviation (0.0063), therefore, Scheme 3 is selected. The node association strength adjustment ratio (increase by 25%), association method optimization suggestions (using a data caching mechanism), and the strategy for adjusting the connection relationship of obstructing nodes (changing the connection protocol) are extracted from Scheme 3 to form the node association adjustment parameters.

[0161] Step S148: Construct multiple resource redistribution schemes. Each resource redistribution scheme includes adjustments to the allocation ratio of unbalanced resource types, resource transfer strategies between node clusters, and resource supplementation configuration suggestions.

[0162] Multiple resource reallocation schemes were constructed, such as Scheme A: The allocation ratio of the unbalanced resource type (CPU) was adjusted to 55% for the first cluster (CG-XZ-002, CG-HT-003) and 35% for the second cluster (CG-RK-005, CG-FK-006). The resource transfer strategy was to transfer 15% of the CPU resources from the first cluster to the second cluster, and the resource supplementation configuration was recommended to supplement the first cluster with an additional 5% of CPU resources. Scheme B: The allocation ratio was adjusted to 50%:40%, the transfer strategy was to transfer 10% of the resources, and the supplementation configuration was recommended to supplement the second cluster with 5% of the resources. Scheme C: The allocation ratio was adjusted to 60%:30%, the transfer strategy was to transfer 5% of the resources, and the supplementation configuration was recommended not to supplement resources, etc. A total of 4 schemes were constructed.

[0163] Step S149: Input each resource reallocation scheme into the complete simulation adaptation scenario, run the simulation process, record the change data of resource allocation imbalance characteristics during the simulation, select a scheme from multiple resource reallocation schemes according to the preset screening rules, and determine the allocation ratio adjustment, transfer instructions and configuration parameters contained therein as resource reallocation parameters.

[0164] Similar to step S147, each resource redistribution scheme is imported into the simulation scenario, the simulation is run, and data such as changes in resource allocation difference values ​​and reduction in the scope of influence are recorded. The resource allocation imbalance mitigation rate is calculated. After multiple rounds of simulation and screening, an optimal scheme is finally determined, and its allocation ratio adjustment, transfer instructions and configuration parameters are extracted to form resource redistribution parameters.

[0165] Step S1410: Construct multiple sets of link evolution guidance schemes. Each set of link evolution guidance schemes includes a stuttering node optimization strategy, a link evolution rate improvement measure, and a stuttering propagation path blocking method. Input each scheme into a complete simulation adaptation scenario, run the simulation process, and record the change data of link evolution stuttering characteristics during the simulation. According to the preset screening rules, select one scheme from the multiple sets of link evolution guidance schemes, and determine the stuttering node adjustment parameters, link evolution rate adjustment instructions, and stuttering propagation path blocking settings contained in it as link evolution guidance parameters. Integrate the node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters to form adaptation optimization parameters.

[0166] Multiple link evolution guidance schemes were constructed. Scheme A: The optimization strategy for the lag node (CG-HT-003) was to increase memory resources; the link evolution rate improvement measure was to optimize the data processing algorithm; and the lag propagation path blocking method was to set up data buffers before and after the lag node. Scheme B: The optimization strategy for the lag node was to optimize code logic; the rate improvement measure was to increase parallel processing threads; and the propagation path blocking method was to limit the data input rate of the lag node. Each scheme was input into a simulation scenario, and the simulation was run. Changes in lag start time offset, lag duration reduction, and propagation path shortening were recorded. The link evolution lag mitigation rate was calculated, the optimal scheme was selected, and relevant parameters were extracted to form link evolution guidance parameters. Finally, the node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters were integrated to form adaptation optimization parameters.

[0167] Step S150: Based on the adaptation optimization parameters, and in conjunction with the real-time status of the dynamic evolution association network, generate a dynamic adaptation business process optimization scheme. The dynamic adaptation business process optimization scheme includes process node association optimization strategy, resource configuration adjustment strategy, and link evolution guidance strategy.

[0168] After obtaining the adaptation and optimization parameters, and combining them with the real-time status of the dynamically evolving network, such as the current association strength of each node, resource consumption, and link evolution trends, an optimization scheme that can dynamically adapt to changes in business processes is formulated. This optimization scheme will propose specific optimization strategies for three aspects: node association, resource configuration, and link evolution, in order to solve bottleneck problems and improve business process efficiency.

[0169] For example, step S151: extract the node association adjustment parameters from the adaptation optimization parameters, and extract the node association strength adjustment ratio, association mode optimization suggestions, and blocking node pair connection relationship adjustment strategies.

[0170] Extract node association adjustment parameters from the adaptation optimization parameters, including the node association strength adjustment ratio (e.g., increase by 25%), association method optimization suggestions (e.g., adopt a data caching mechanism), and strategies to block node pair connection relationship adjustment (e.g., change the connection protocol).

[0171] Step S152: Extract the real-time state data of the dynamic evolution association network, and obtain the real-time changes of the association strength value, connection relationship type and node association hindrance characteristics of each process node.

[0172] Real-time status data is obtained from the dynamic evolutionary association network, such as the current association strength value of CG-XZ-002 and CG-HT-003 being 0.6, the connection relationship type being serial connection, and the real-time change of the node association blockage characteristics, such as the blockage duration being shortened to 1 hour.

[0173] Step S153: Based on the node association adjustment parameters and the real-time status data of the dynamically evolving association network, formulate a process node association optimization strategy. The process node association optimization strategy includes specific values ​​for adjusting the association strength of each node pair, steps for changing the association method, and a scheme for reconstructing the connection relationship of the blocking node pairs.

[0174] Based on the node association strength adjustment ratio (increase by 25%) and the current association strength value (0.6), the adjusted value is calculated to be 0.6 × (1 + 25%) = 0.75. The steps for changing the association method are: first, enable the data caching module in the system, configure the cache size and update frequency, then test the impact of the caching mechanism on data transmission, and finally officially switch to the cached transmission mode. The solution for reconstructing the connection relationship of obstructing nodes is: disable the original TCP connection protocol, enable the UDP connection protocol, and configure packet verification and retransmission mechanisms to improve connection stability and transmission efficiency. Integrating these elements forms the process node association optimization strategy.

[0175] Step S154: Extract the resource redistribution parameters from the adaptation optimization parameters, and extract the allocation ratio adjustment of unbalanced resource types, resource transfer strategies between node clusters, and resource supplementation configuration suggestions.

[0176] Extract resource redistribution parameters from the adaptation and optimization parameters. For example, adjust the allocation ratio of unbalanced resource type (CPU) to 55% for the first cluster and 35% for the second cluster. The resource transfer strategy is to transfer 15% of CPU resources from the first cluster to the second cluster. The resource supplementation configuration suggestion is to supplement the first cluster with 5% additional CPU resources.

[0177] Step S155: Combining the node resource occupancy correlation data in the real-time state of the dynamically evolving network with the real-time resource requirements of the business process, formulate a resource allocation adjustment strategy. The resource allocation adjustment strategy includes the specific allocation ratio of each unbalanced resource type, the time node for resource transfer, and the configuration standard for supplementary resources.

[0178] The real-time node resource usage data in the dynamically evolving network shows that the first cluster currently uses 70% of its CPU, while the second cluster uses 20%. The real-time resource requirement of the business process is that the inbound acceptance stage (belonging to the second cluster) needs more CPU resources to accelerate processing. Based on this, a resource allocation adjustment strategy is formulated: the specific allocation ratio for each unbalanced resource type (CPU) is 55% for the first cluster and 35% for the second cluster; the resource transfer time is chosen to be during off-peak business hours (e.g., 8 PM) to minimize the impact on ongoing business; the configuration standard for supplementing resources is that the supplemented 5% CPU resources must have a clock speed of no less than 3.0 GHz and a latency of no more than 5 ms.

[0179] Step S156: Extract the link evolution guidance parameters from the adaptation optimization parameters, and extract the lag node optimization strategy, link evolution rate improvement measures, and lag propagation path blocking method.

[0180] The link evolution guidance parameters are extracted from the adaptation optimization parameters. For example, the optimization strategy for the stuttering node (CG-HT-003) is to increase memory resources (e.g., increase 4GB of memory), the link evolution rate improvement measure is to optimize the data processing algorithm (e.g., adopt the divide-and-conquer algorithm), and the stuttering propagation path blocking method is to set up a data buffer (buffer size is 100MB) before and after the stuttering node.

[0181] Step S157: Based on the link evolution trend data and network structure adaptability characteristics in the real-time state of the dynamically evolving network, formulate a link evolution guidance strategy. The link evolution guidance strategy includes specific optimization steps for stuck nodes, phased improvement targets for link evolution rate, and setting blocking nodes for stuck propagation paths.

[0182] The link evolution trend data in the real-time status of the dynamically evolving network shows that the current link evolution rate is 0.03 / hour, and the network structure adaptability characteristic value is 0.5. A link evolution guidance strategy is formulated: the specific optimization steps for the stuck node are as follows: first, stop the non-critical tasks of the stuck node to release memory resources; then, add 4GB of physical memory; and finally, restart the node service. The phased improvement target for the link evolution rate is: to increase to 0.04 / hour in the first phase (1-2 hours), to increase to 0.05 / hour in the second phase (3-4 hours), and to stabilize at above 0.05 / hour in the third phase (after 5 hours). The blocking nodes for the stuck propagation path are set up with data buffers between CG-XZ-002 and CG-HT-003, and between CG-HT-003 and CG-DD-004, with a buffer size of 100MB. When a stuck sign is detected, the buffers are automatically activated to absorb data fluctuations.

[0183] Step S158: Integrate process node association optimization strategies, resource allocation adjustment strategies, and link evolution guidance strategies to form a preliminary dynamic adaptable business process optimization solution.

[0184] The process node association optimization strategy formulated in step S153, the resource configuration adjustment strategy formulated in step S155, and the link evolution guidance strategy formulated in step S157 are integrated to form a preliminary dynamic adaptive business process optimization scheme. This dynamic adaptive business process optimization scheme covers optimization content in three aspects: node association, resource configuration, and link evolution.

[0185] Step S159: Verify the conformity of the initial dynamic adaptive business process optimization plan with the core execution objectives of the business process, and revise the strategy content based on the verification results.

[0186] The core execution objectives of the business process are to improve procurement efficiency, reduce procurement costs, and ensure procurement quality. The initial optimization plan is validated against these objectives. For example, it checks whether process node association optimization strategies can improve data flow efficiency, whether resource allocation adjustment strategies can reduce resource waste and thus lower costs, and whether link evolution guidance strategies can ensure the stability of the business process to guarantee procurement quality. If the validation finds that resource supplementation suggestions may increase costs and exceed the budget, they are revised, such as adjusting the supplementation of 5% CPU resources to 3%, and resource utilization is improved by optimizing the resource scheduling algorithm to achieve a balance between cost and efficiency.

[0187] Step S1510: In the dynamic adaptive business process optimization scheme, monitoring indicators and dynamic adjustment rules are set. The monitoring indicators include the correlation strength optimization compliance rate, resource configuration adjustment fit rate, and link evolution guidance effectiveness. The dynamic adjustment rules define the logic and steps for triggering strategy correction based on the monitoring indicators, thus forming the final dynamic adaptive business process optimization scheme.

[0188] Set monitoring metrics: The correlation strength optimization compliance rate is the percentage of actual correlation strength reaching the target value, where the target value is the adjusted correlation strength value (e.g., 0.75); the resource allocation adjustment fit rate is the degree of conformity between the actual resource allocation ratio and the target allocation ratio; the link evolution guidance effectiveness rate is the percentage of link evolution rate reaching the target rate. Dynamic adjustment rules include: when the correlation strength optimization compliance rate is below 90% for three consecutive time windows, trigger a reassessment and adjustment of the node correlation strength adjustment ratio; when the resource allocation adjustment fit rate is below 85%, check the resource transfer strategy and supplementary configuration standards, and make corresponding corrections; when the link evolution guidance effectiveness rate is below 80%, reanalyze the characteristics of stuck nodes and adjust the optimization strategy. Incorporate the above monitoring metrics and dynamic adjustment rules into the optimization plan to form the final dynamically adaptable business process optimization plan.

[0189] Figure 2 The following is a schematic diagram of the hardware structure of the intelligent business process analysis system 100 for multimodal data provided in an embodiment of the present invention. Figure 2 As shown, the intelligent business process analysis system 100 for multimodal data may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0190] Machine-readable storage medium 120 can store data and / or instructions. In some embodiments, machine-readable storage medium 120 can store data acquired from an external terminal. In some embodiments, machine-readable storage medium 120 can store data and / or instructions used by the intelligent business process analysis system 100 for multimodal data to execute or use in order to complete the exemplary methods described in this invention. In a specific implementation, one or more processors 110 execute the computer-executable instructions stored in machine-readable storage medium 120, enabling processor 110 to execute the intelligent business process analysis method for multimodal data as described in the above method embodiments. Processor 110, machine-readable storage medium 120, and communication unit 140 are connected via bus 130, and processor 110 can be used to control the sending and receiving actions of communication unit 140. The specific implementation process of processor 110 can be found in the various method embodiments executed by the intelligent business process analysis system 100 for multimodal data described above, and their implementation principles and technical effects are similar, so they will not be repeated here.

[0191] Furthermore, embodiments of the present invention also provide a readable storage medium containing computer-executable instructions. When a processor executes the computer-executable instructions, the above-described intelligent business process analysis method for multimodal data is implemented.

[0192] It should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof.

Claims

1. A method for intelligent business process analysis of multimodal data, characterized in that, The method includes: Collect multimodal process data generated during the operation of business processes, perform process node link mapping processing on the multimodal process data, and generate a process node association matrix. The multimodal process data includes system log data, user operation behavior data, document flow data, and collaborative interaction data. The process node association matrix includes the connection relationship, data flow intensity, and resource consumption association of each process node. Based on the process node association matrix and the real-time running status of the business process, the process node association network is dynamically evolved to generate a dynamically evolving association network. The dynamically evolving association network includes the dynamic change trajectory of node association strength, link evolution trend and network structure adaptability characteristics. Bottleneck gene localization processing is performed on the dynamic evolutionary association network to extract bottleneck gene features in the process node association. The bottleneck gene features include node association obstruction features, resource allocation imbalance features, and link evolution stagnation features. A simulation adaptation scenario is constructed, and the bottleneck gene features are input into the simulation adaptation scenario for dynamic deduction and optimization to generate adaptation optimization parameters. The adaptation optimization parameters include node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters. Based on the aforementioned adaptation and optimization parameters, and in conjunction with the real-time status of the dynamically evolving network, a dynamically adaptable business process optimization scheme is generated. The dynamically adaptable business process optimization scheme includes a process node association optimization strategy, a resource configuration adjustment strategy, and a link evolution guidance strategy. The process node association network is dynamically evolved based on the process node association matrix and the real-time running status of the business process to generate a dynamically evolving association network, including: The initial process node association network is constructed by using the process node identifiers in the process node association matrix as nodes, the connection relationships as edges, and the data flow intensity and resource consumption association as edge attributes. Collect real-time operational status data of business processes, including real-time execution progress, real-time resource occupancy, and real-time data flow rate of each process node. Extract the difference between the real-time execution progress and the historical execution progress of each process node from the real-time running status data, and generate node progress deviation features. The node progress deviation features are generated based on the degree of deviation between the real-time progress and the historical average progress. Analyze the degree of conformity between the real-time resource occupancy of each process node and the preset resource configuration standard, and generate node resource occupancy deviation characteristics. The node resource occupancy deviation characteristics are generated based on the difference between the real-time occupancy and the preset configuration. Calculate the ratio of the real-time data flow rate to the historical average flow rate for each process node, and generate node data flow deviation characteristics. The node data flow deviation characteristics are generated based on the ratio between the real-time rate and the historical rate. Integrate node progress deviation characteristics, node resource occupancy deviation characteristics, and node data flow deviation characteristics to form a node operation status deviation set; based on the node operation status deviation set, process it through a preset driving analysis model to generate network evolution driving factors, which are parameters used to comprehensively reflect the degree of influence of node state changes on network correlation strength. Input the network evolution driving factor into the initial process node association network, adjust the attribute values ​​of the edges in the initial process node association network, update the node association strength in the initial process node association network, and generate the process node association network after the first evolution. Based on a preset evolution time window, the process node association network sequence is formed by repeatedly collecting real-time running status data, generating network evolution driving factors, and adjusting the network attributes associated with process nodes. Extract the change data of the association strength of each node in the process node association network sequence, and generate the dynamic change trajectory of the node association strength. The dynamic change trajectory of the node association strength is a continuous change curve of the association strength under each time window. Analyze the changes in the link structure of the process node association network sequence, generate the link evolution trend, and calculate the degree of fit between the process node association network structure and business process requirements at each evolution stage to generate network structure adaptability characteristics. Integrate the dynamic change trajectory of node association strength, link evolution trend and network structure adaptability characteristics to form a dynamic evolution association network. The bottleneck gene localization process for the dynamic evolutionary association network, extracting bottleneck gene features from the process node associations, includes: Extract the dynamic change trajectory of node association strength in the dynamic evolution association network, and filter out node pairs whose association strength continuously decreases. The node pair is a combination of two process node identifiers with a direct connection relationship. Analyze the link evolution trend corresponding to node pairs with continuously decreasing association strength, and determine the stagnation period of link evolution. The stagnation period is a continuous time interval in which the rate of decrease in association strength exceeds a preset threshold. Based on the node progress deviation characteristics and node data flow deviation characteristics during the stagnation period, node-related blockage characteristics are generated. The node-related blockage characteristics include the blockage start time, blockage duration, and blockage impact range. Extract resource usage correlation data of each process node in the dynamic evolution correlation network, analyze the changes in the allocation ratio of the same resource type among different process nodes, and filter out resource types whose allocation ratio fluctuations exceed a preset range; Based on the node resource usage deviation characteristics corresponding to the resource type, the first process node cluster with a real-time resource usage that is consistently higher than the preset configuration amount and the second process node cluster with a real-time resource usage that is consistently lower than the preset configuration amount are identified. Based on the resource consumption of the first process node cluster and the second process node cluster, the resource allocation difference between them is calculated; combined with the connection relationship of the associated node pairs connected to the first process node cluster and the second process node cluster and the data flow intensity, the impact range of the resource allocation difference is evaluated, and a resource allocation imbalance feature is generated. The resource allocation imbalance feature includes the imbalanced resource type, the identifier of the first process node cluster and the second process node cluster, the resource allocation difference value, and the impact range evaluation result. Extract the link evolution trend data in the dynamic evolution association network, and filter the links whose evolution rate is consistently lower than the preset evolution rate. The links are continuous association paths connecting multiple process nodes. Analyze the dynamic change trajectory of the node association strength and the network structure adaptability characteristics of the link to identify the key nodes that cause the link evolution to stall. The key nodes are the core process nodes in the link that cause the evolution rate to decrease. Based on the progress deviation characteristics, resource usage deviation characteristics, and data flow deviation characteristics of key nodes, link evolution stuttering characteristics are generated. The link evolution stuttering characteristics include stuttering node identifier, stuttering start time period, and stuttering propagation path. By integrating the characteristics of node association blockage, resource allocation imbalance, and link evolution stagnation, bottleneck gene characteristics in process node association are formed.

2. The intelligent business process analysis method for multimodal data according to claim 1, characterized in that, The step of performing process node link mapping processing on the multimodal process data to generate a process node association matrix includes: Extract the process node identifiers corresponding to each type of data in the multimodal process data. The process node identifier is a unique identifier that represents an independent execution link in the business process, and each process node identifier corresponds to a specific business execution action. Analyze the execution time series corresponding to the process node identifiers in the system log data, and generate the time distribution characteristics of each process node identifier. The time distribution characteristics include the execution start time, the execution duration, and the execution interval period. Extract the sequence of operation actions corresponding to each process node identifier from the user operation behavior data, and generate the operation association feature for each process node identifier. The operation association feature includes the operation triggering condition, operation execution order and operation result feedback. The document transfer records corresponding to each process node identifier in the document flow data are parsed to generate document association features for each process node identifier. The document association features include document type, transfer path and transfer time. Mine the associations of collaborative entities corresponding to each process node identifier in the collaborative interaction data, and generate collaborative association features for each process node identifier. The collaborative association features include collaborative entity type, interaction frequency and interaction content association. Integrate the time distribution characteristics, operation association characteristics, document association characteristics, and collaboration association characteristics of each process node identifier to form a comprehensive association feature set for each process node identifier; Based on the comprehensive associated feature set of all process node identifiers, the feature similarity between any two process node identifiers is calculated. The feature similarity is generated based on the matching degree of each associated feature. Based on feature similarity, establish a connection between any two process node identifiers. The connection is determined based on the degree of conformity of feature similarity. Calculate the total data flow and resource usage ratio corresponding to each connection relationship, and generate data flow intensity and resource usage association. Data flow intensity is generated based on the total data transmission volume, and resource usage association is generated based on the usage ratio of shared resources. A process node association matrix is ​​constructed using process node identifiers as matrix rows and columns, and connection relationships, data flow intensity, and resource consumption associations as matrix elements.

3. The intelligent business process analysis method for multimodal data according to claim 1, characterized in that, The construction of the simulation adaptation scenario involves inputting the bottleneck gene features into the simulation adaptation scenario for dynamic deduction and optimization, generating adaptation optimization parameters, including: Based on the core execution objectives and preset operating standards of the business process, a basic simulation scenario is constructed. The basic simulation scenario includes the basic configuration of process nodes, the basic resource allocation scheme, and the basic link connection mode. Extract the node association blockage features from the bottleneck gene features, determine the blockage node pairs and the blockage influence range, import the blockage node pairs, the blockage influence range and the corresponding blockage feature parameters into the basic simulation scenario, and generate a simulation sub-scenario containing the blockage problem. Extract the resource allocation imbalance features from the bottleneck gene features, determine the imbalanced resource type, imbalanced node cluster identifier and imbalance impact degree, import the corresponding information into the basic simulation scenario, and generate a simulation sub-scenario containing the resource imbalance problem. Extract the link evolution stuttering features from the bottleneck gene features, determine the stuttering node identifier, stuttering start time and stuttering propagation path, import the corresponding information into the basic simulation scenario, and generate a simulation sub-scenario containing the link stuttering problem. Integrate simulation sub-scenarios containing bottleneck issues, simulation sub-scenarios containing resource imbalance issues, and simulation sub-scenarios containing link blockage issues, and overlay the impact and correlation of each issue to generate a complete simulation adaptation scenario. The complete simulation adaptation scenario includes all bottleneck issues in the business process and the interaction relationships between issues. Construct multiple sets of node association adjustment schemes. Each set of node association adjustment schemes includes the node association strength adjustment ratio, association method optimization suggestions, and strategies for adjusting the connection relationship of blocking node pairs. Input each set of node association adjustment schemes into the complete simulation adaptation scenario, run the simulation process, record the change data of node association hindrance characteristics during the simulation, and select a set of schemes from multiple sets of node association adjustment schemes according to the preset screening rules. The node association strength adjustment ratio, association mode adjustment parameters and hindrance node pair connection relationship adjustment instructions contained in the scheme are determined as node association adjustment parameters. Construct multiple resource reallocation schemes, each of which includes adjustments to the allocation ratio of imbalanced resource types, resource transfer strategies between node clusters, and resource supplementation configuration suggestions. Each resource reallocation scheme is input into the complete simulation adaptation scenario, the simulation process is run, and the changes in the resource allocation imbalance characteristics during the simulation are recorded. According to the preset filtering rules, a scheme is selected from multiple resource reallocation schemes, and the allocation ratio adjustment, transfer instructions and configuration parameters contained therein are determined as resource reallocation parameters. Multiple link evolution guidance schemes are constructed. Each scheme includes a stuttering node optimization strategy, a link evolution rate improvement measure, and a stuttering propagation path blocking method. Each scheme is input into a complete simulation adaptation scenario, the simulation process is run, and the change data of link evolution stuttering characteristics during the simulation are recorded. According to the preset screening rules, one scheme is selected from the multiple link evolution guidance schemes, and the stuttering node adjustment parameters, link evolution rate adjustment instructions, and stuttering propagation path blocking settings contained in it are determined as link evolution guidance parameters. The node association adjustment parameters, resource reallocation parameters, and link evolution guidance parameters are integrated to form adaptation optimization parameters.

4. The intelligent business process analysis method for multimodal data according to claim 2, characterized in that, The comprehensive associated feature set based on all process node identifiers calculates the feature similarity between any two process node identifiers. The feature similarity is generated comprehensively based on the matching degree of each associated feature, including: Extract the time distribution features of any two process node identifiers, calculate the similarity indices of execution start time difference, execution duration overlap ratio and execution interval period fit, and combine the three similarity indices to generate time dimension evaluation results. Extract the operation association features of any two process node identifiers, calculate the similarity index of operation trigger condition similarity, operation execution order fit and operation result feedback association respectively, and combine the three similarity indexes to generate operation dimension evaluation results. Extract the document association features of any two process node identifiers, calculate the similarity indexes of document type matching degree, transmission path overlap ratio and transmission timeliness fit, and combine the three similarity indexes to generate document dimension evaluation results. Extract the collaboration association features of any two process node identifiers, calculate the similarity index of collaboration subject type similarity, interaction frequency association and interaction content matching degree respectively, and combine the three similarity indexes to generate collaboration dimension evaluation results. Extract data flow records and resource sharing records related to the two process node identifiers from the multimodal process data, calculate the data flow frequency correlation similarity index and the resource sharing depth similarity index, and comprehensively generate data and resource dimension evaluation results; Based on the evaluation results of time dimension, operation dimension, document dimension, collaboration dimension, and data and resource dimension, a comprehensive analysis is performed using preset association rules to obtain the feature similarity between any two process node identifiers. This completes the feature similarity calculation between all any two process node identifiers, forming a feature similarity matrix.

5. The intelligent business process analysis method for multimodal data according to claim 1, characterized in that, The process of inputting network evolution driving factors into the initial process node association network, adjusting the attribute values ​​of the edges in the initial process node association network, updating the node association strength in the initial process node association network, and generating the process node association network after the first evolution includes: Extract the specific values ​​of node progress deviation characteristics, node resource utilization deviation characteristics, and node data flow deviation characteristics from the network evolution driving factors; Analyze the impact of node schedule deviation characteristic values ​​on the connection relationships of corresponding process nodes, and determine the edge attribute adjustment ratio of the initial process node association network corresponding to the node schedule deviation characteristic. The adjustment ratio is generated based on the magnitude of the deviation characteristic value. Based on the node resource usage deviation characteristic value, the adjustment range of the corresponding resource usage association attribute is determined. The adjustment range is generated based on the ratio between the deviation characteristic value and the preset deviation threshold. Based on the numerical value of the deviation feature of node data flow, the adjustment coefficient of the corresponding data flow intensity attribute is determined. The adjustment coefficient is generated based on the sign and magnitude of the deviation feature value. The adjustment ratio, adjustment range, and adjustment coefficient are integrated into a set of edge attribute adjustment parameters, which includes the adjustment parameters of each attribute of the edge of the network associated with each initial process node. Traverse all edges in the network associated with the initial process nodes and locate the process node identifier combination corresponding to each edge in the network associated with the initial process nodes. Based on the combination of process node identifiers, match the corresponding adjustment parameters from the set of edge attribute adjustment parameters; Modify the data flow intensity attribute value and resource consumption association attribute value of the edges of the network associated with each initial process node according to the adjustment parameters; Based on the adjusted edge attribute values, the node association strength between the two process node identifiers connected by the edge of each initial process node association network is recalculated. The node association strength is generated comprehensively based on the adjusted attribute values. Update all edge attribute values ​​and node association strengths in the initial process node association network to form the process node association network after the first evolution.

6. The intelligent business process analysis method for multimodal data according to claim 1, characterized in that, Based on the node progress deviation characteristics and node data flow deviation characteristics during the stagnation period, node-related stall characteristics are generated. These characteristics include the stall start time, stall duration, and stall impact range, including: Extract the start and end times of the stagnation period, and determine the start time as the stagnation initiation time. Calculate the time difference between the end time and the start time, and determine the duration of the blockade. Extract the process node identifiers corresponding to node pairs whose association strength continuously decreases during the stagnation period, and record the two process node identifiers for each node pair. Analyze the number of connected nodes and the length of the connection links for each process node in the dynamically evolving association network to determine the association radiation range of each node; Based on the associated radiation range of each process node identifier in the node pair, calculate the intersection and union regions of the two radiation ranges; Extract all process node identifiers within the intersection area and determine the business execution stage corresponding to the process node identifier; Analyze the logical correlation between the business execution links and nodes and the corresponding business execution links, and filter process node identifiers that meet the preset standards for logical correlation. Integrate the business execution steps corresponding to the filtered process node identifiers to determine the coverage of the business execution steps; By combining the decrease in data flow intensity during the duration of the blockage, the depth of the impact of the blockage on the business execution process within the coverage area can be determined. Based on the start time of the blockade, the duration of the blockade, the coverage area, and the depth of influence, node-related blockade features are generated.

7. The intelligent business process analysis method for multimodal data according to claim 3, characterized in that, The process involves inputting each set of node association adjustment schemes into a complete simulation adaptation scenario, running the simulation process, recording the changes in node association hindrance characteristics during the simulation, and selecting the scheme that best alleviates the hindrance characteristics as the node association adjustment parameters. This includes: Import the first set of node association adjustment schemes into the complete simulation adaptation scenario, set the simulation run time step and total duration. The time step is set based on the execution cycle of the business process, and the total duration covers a complete business process execution cycle. Start the simulation process and record the changes in various parameters of the node-related stall characteristics during the simulation at the set time step, including the offset of the stall start time, the reduction of the stall duration, and the reduction ratio of the stall influence range. After the simulation process is completed, the mitigation effects of the blockage onset time, blockage duration and blockage range are evaluated respectively, and the blockage feature mitigation rate is calculated based on the comprehensive mitigation effects of each factor. Following the same simulation settings and recording logic, the remaining node association adjustment schemes were sequentially imported into the complete simulation adaptation scenario, the simulation process was run and the data on the change of stall characteristics were recorded, and the stall characteristic mitigation rate of each scheme was calculated. Extract the blockage relief rate data of all group schemes, sort them in descending order of relief rate value, and select the top K schemes as candidate optimization schemes. Rerun the simulation process of the candidate optimization scheme, increase the number of simulation runs, and obtain the average mitigation rate of the stall characteristics after multiple rounds of simulation; Calculate the standard deviation of the mitigation rate of the candidate optimization schemes in multiple rounds of simulation, wherein the standard deviation of the mitigation rate is generated based on the degree of deviation of the mitigation rate from the average value in multiple rounds; Based on the average and standard deviation of the mitigation rate of the blocking feature after multiple rounds of simulation, a set of schemes is determined from the candidate optimization schemes according to the preset final selection rules. The node association strength adjustment ratio, association mode optimization suggestions and blocking node pair connection relationship adjustment strategies are extracted from the scheme to form node association adjustment parameters.

8. An intelligent business process analysis system for multimodal data, characterized in that, The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to run the programs, instructions, or code in the memory to implement the intelligent business process analysis method for multimodal data as described in any one of claims 1-7.

Images