Processing method and processing device for a workflow

Abstract

The application discloses a processing method and processing device for a workflow. The processing method comprises: obtaining input information; processing the input information based on a knowledge base unit by an intelligent agent unit to generate workflow information corresponding to the input information, the workflow information being executable to realize a function corresponding to the input information; the intelligent agent unit comprising intelligent agent modules for executing different generation tasks, each intelligent agent module generating a generation result corresponding to the generation task thereof, at least one intelligent agent module taking a generation result generated by a previous intelligent agent module thereof as an input result to execute the generation task corresponding thereto; at least part of the intelligent agent modules obtaining knowledge information corresponding to the generation task thereof from the knowledge base unit in a state of executing the generation task corresponding thereto, and executing the generation task based on the knowledge information. The application can improve the generation quality and efficiency of the workflow.

Description

Technical Field

[0001] This application relates to the field of information processing, and more specifically, to a processing method and apparatus for workflow. Background Technology

[0002] Workflow orchestration and generation are core technologies for business automation, widely used in areas such as data synchronization task orchestration, ETL process generation, and batch processing task scheduling. Traditional workflow orchestration methods mainly rely on manual configuration, which suffers from low efficiency, high barriers to entry, and high maintenance costs.

[0003] To achieve automated workflow generation, existing technologies utilize the natural language understanding and generation capabilities of large language models to handle user business needs, but these technologies suffer from low processing efficiency and cannot meet user requirements. Summary of the Invention

[0004] The main objective of this application is to provide a workflow processing method and apparatus to solve the technical problem of over-reliance on a single intelligent agent in the automatic workflow generation process in the prior art, thereby effectively improving the generation quality and efficiency of workflows.

[0005] To achieve the above objectives, a first aspect of this application proposes a workflow processing method, comprising: obtaining input information; processing the input information by an intelligent agent unit based on a knowledge base unit to generate workflow information corresponding to the input information, wherein the workflow information can be executed to achieve a function corresponding to the input information; wherein the intelligent agent unit includes intelligent agent modules for executing different generation tasks, each intelligent agent module generating a generation result corresponding to its generation task, and at least one intelligent agent module using the generation result generated by its previous intelligent agent module as input to execute its corresponding generation task; wherein at least some of the intelligent agent modules, while executing their corresponding generation tasks, obtain knowledge information corresponding to the generation task from the knowledge base unit and execute the generation task based on the knowledge information.

[0006] Optionally, when the intelligent agent module is executing its corresponding generation task, it obtains knowledge information corresponding to the generation task from the knowledge base unit, including: generating knowledge query information based on the type of the generation task corresponding to the intelligent agent module and the input result obtained by the intelligent agent module; determining the knowledge information in the knowledge base unit that matches the knowledge query information; and using the knowledge information as the generation basis and / or constraint basis for the intelligent agent module to execute the generation task.

[0007] Optionally, the intelligent agent unit includes: a first intelligent agent module, configured to parse the input information, obtain a structured description corresponding to the input information, and generate a first generation result based on the structured description; a second intelligent agent module, configured to determine component planning information corresponding to the structured description based on the first generation result, and generate a second generation result including the component planning information, wherein the component planning information is used to characterize the workflow nodes corresponding to the structured description and the association relationships between the workflow nodes; and a third intelligent agent module, configured to determine parameter information corresponding to the component planning information based on the second generation result, and generate a third generation result including the parameter information, wherein the parameter information is used to characterize the execution parameters of the workflow nodes, wherein the workflow information is generated based on the third generation result.

[0008] Optionally, the first intelligent agent module generates the first generation result by: the first intelligent agent module determining first knowledge query information based on the input information; obtaining first knowledge information matching the first knowledge query information from the knowledge base unit; and generating the first generation result based on the first knowledge information and the input information if the matching degree between the first knowledge information and the input information meets a precondition, wherein the first generation result includes at least one of the following combinations: triggering condition, data source, data target, transformation rule, and constraint condition.

[0009] Optionally, the second intelligent agent module generates the second generation result by: generating component planning information based on the first generation result; obtaining second knowledge information from the knowledge base unit, wherein the second knowledge information matches the component planning information or matches the first generation result; and generating the second generation result based on the component planning information and the second knowledge information, wherein the second generation result includes at least one of the following combinations: a selected list of components, inter-component dependencies, and reasons for component selection.

[0010] Optionally, the third intelligent agent module generates the third generation result by: generating parameter information based on the first generation result and the second generation result; verifying the parameter information based on the parameter constraint rules in the knowledge base unit; optimizing the parameter information based on the knowledge base unit in response to the parameter information passing the verification; and generating the third generation result based on the second generation result and the optimized parameter information.

[0011] Optionally, the method further includes: obtaining execution feedback information during the execution of the workflow information; if, based on the knowledge base unit, it is determined that the execution feedback information meets preset optimization conditions, then generating an optimization strategy for adjusting the workflow information, and generating optimized workflow information based on the optimization strategy.

[0012] Optionally, the method further includes: in response to the execution feedback information satisfying preset optimization conditions, at least inputting the execution feedback information into a large language model; and obtaining an optimization strategy for the workflow information output by the large language model.

[0013] Optionally, the method further includes: the intelligent agent modules transmitting the generation result through a target message protocol, wherein the message structure of the target message protocol includes at least: the generation result and confirmation information, and / or obtaining the execution result of the workflow information in response to the execution of the workflow information; and storing some or all information related to the workflow information in the knowledge base unit based on the execution result of the workflow information.

[0014] A second aspect of this application provides a workflow processing apparatus, comprising: an acquisition unit for acquiring input information; a knowledge base unit including knowledge information; and an intelligent agent unit for processing the input information based on the knowledge base unit to generate workflow information corresponding to the input information, wherein the workflow information can be executed to achieve a function corresponding to the input information; wherein the intelligent agent unit includes intelligent agent modules for executing different generation tasks, each intelligent agent module generating a generation result corresponding to its generation task, and at least one intelligent agent module using the generation result generated by its previous intelligent agent module as input to execute its corresponding generation task; wherein at least some of the intelligent agent modules, while executing their corresponding generation tasks, obtain knowledge information corresponding to the generation task from the knowledge base unit and execute the generation task based on the knowledge information.

[0015] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the workflow processing method described in the first aspect or any implementation thereof.

[0016] A fourth aspect of this application provides an electronic device including at least one processor and a memory connected to the processor, wherein the memory is used to store a computer program; the processor is used to execute the computer program to enable the electronic device to implement the workflow processing method described in the first aspect or any implementation thereof.

[0017] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the workflow processing method described in the first aspect or any implementation thereof.

[0018] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In this application, multi-agent modules collaborate to execute generation tasks, allowing specialized agent modules to handle specialized generation tasks, effectively avoiding decision fatigue. Furthermore, during the generation process, agent modules can match the required knowledge information based on knowledge base units, effectively improving workflow generation efficiency and enhancing the consistency and stability of workflow information generated from the same or similar input information. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart of a workflow processing method provided in this application; Figure 2 A schematic diagram of the architecture of an intelligent agent unit provided in this application; Figure 3 An architectural diagram of the workflow processing apparatus provided in this application; Figure 4 A flowchart of the workflow processing apparatus provided in this application; Figure 5 A schematic diagram illustrating the working principle of the adaptive optimization engine provided in this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0023] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0024] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linked," and "socketing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0025] To address one or more technical problems existing in the prior art, embodiments of this application provide a workflow processing method. Please refer to [link to previous document]. Figure 1 Here is a flowchart of the method, which may include the following steps.

[0026] S101, obtain input information.

[0027] The “input information” in this application may be at least one of the following: voice information, natural language text, form information, configuration fragments, interface requests, semi-structured data, task descriptions, graphical information, and link information.

[0028] This solution can be applied to various scenarios, such as data synchronization, data processing, business process automation, automated operation and maintenance, interface orchestration, and task scheduling. The input information can take different forms depending on the scenario or input method.

[0029] For example, the input information could be a business requirement described in natural language: synchronize the order table from MySQL to Hive every morning at 8:00 AM.

[0030] For example, in business process automation scenarios, input information can be process triggering conditions, data source information, target system information, etc., filled in on forms. In automated operations and maintenance scenarios, input information can be interface requests submitted by operations and maintenance personnel, including descriptions of the operations and maintenance tasks to be executed. In interface orchestration scenarios, input information can be semi-structured descriptions of interface call sequences. In task scheduling scenarios, input information can be logs or descriptive information of historical task executions.

[0031] S102, the intelligent agent unit processes the input information based on the knowledge base unit to generate workflow information corresponding to the input information.

[0032] In this application, "workflow information" can refer to information that can be recognized and executed by program modules such as workflow engines, scheduling systems, task execution systems, automated orchestration systems, and script execution environments. Specifically, workflow information can be executed to achieve functions corresponding to the input information.

[0033] The intelligent agent unit includes intelligent agent modules for performing different generation tasks. Each intelligent agent module generates a generation result corresponding to its generation task, and at least one intelligent agent module can use the generation result generated by the preceding intelligent agent module as input to execute its corresponding generation task. Through this multi-agent module collaboration, the generation result of the preceding intelligent agent module can form a clear input boundary for the subsequent intelligent agent module, enabling multiple intelligent agent modules to cooperate and allowing specialized intelligent agent modules to handle specialized tasks, thus avoiding decision fatigue.

[0034] The "intelligent agent module" in this application can be implemented by one or more of the following: a large language model, a machine learning model, a rule engine, a program module, a model and rule combination module, and other modules with input processing and result generation capabilities. The specific module can be selected and configured based on the application scenario and usage requirements of this application.

[0035] In this implementation, different intelligent agent modules can run on the same computing device or be distributed across different computing devices or service nodes. For example, in one embodiment, all intelligent agent modules are deployed on the same server and interact with each other via inter-process communication or inter-thread communication. In another embodiment, different intelligent agent modules can be deployed on different servers or cloud service nodes and communicate via remote procedure calls (RPC), message queues, or HTTP interfaces. In yet another embodiment, some intelligent agent modules can be deployed on edge computing nodes, while others are deployed in cloud data centers to achieve more flexible resource scheduling and load balancing.

[0036] Knowledge base units can be pre-built before executing S102. For example, knowledge base units can be built based on execution files corresponding to historical cases, shared case files, and / or manually entered information.

[0037] In some alternative implementations, the knowledge base units can also be dynamically updated and expanded.

[0038] At least some agent modules can obtain knowledge information corresponding to their generation tasks from the knowledge base unit when performing generation tasks, and execute the generation tasks based on the knowledge information. In other words, during the execution of generation tasks, agent modules do not rely entirely on their own model capabilities or program logic, but actively query the knowledge base unit to obtain knowledge information related to the current generation task, and use this knowledge information as the basis, reference or constraint for executing the generation task.

[0039] This application proposes a more granular division of labor among the various stages of the workflow generation process based on input information. Through collaborative generation by multiple intelligent agent modules, specialized intelligent agent modules can handle specialized generation tasks, effectively avoiding decision fatigue. Furthermore, during the execution of generation tasks, intelligent agent modules can match the required knowledge information based on knowledge base units, effectively improving the efficiency of workflow information generation and the consistency of generated results.

[0040] To address the need for automatically generated workflow information, an existing technology is based on a single-agent architecture, which gradually builds the workflow through an iterative "think-act-observe" cycle. Specifically, a single agent analyzes user needs, selects appropriate components, and then gradually constructs the complete workflow structure. While this architecture is simple and easy to implement, it suffers from decreased decision-making quality when handling complex workflow requirements. Furthermore, as the complexity of the workflow requirements increases, the quality of the generated workflow information deteriorates, and processing time is significantly extended.

[0041] The technical solution provided in this application can effectively solve the problems of efficiency and quality in generating workflow information.

[0042] Figure 2 This is a schematic diagram of the architecture of an intelligent agent unit provided in an embodiment of this application. See also... Figure 2 The intelligent agent unit may specifically include a first intelligent agent module 201, a second intelligent agent module 202, and a third intelligent agent module 203.

[0043] The first intelligent agent module 201 is used to parse the input information, obtain the structured description corresponding to the input information, and generate the first generation result based on the structured description.

[0044] The specific structure of the structured description can be defined by the user based on the business scenario, or it can be set to a fixed structure. For example, the specific structure of the structured description may include: trigger conditions, data source, data target, transformation rules, and constraints.

[0045] As an example, the first intelligent agent module 201 can be a requirement analysis intelligent agent that can receive natural language requirements input by the user. For example, if the input information of the first intelligent agent module 201 is "synchronize the order table from MySQL to Hive at 8 am every day", then the corresponding structured output requirements include: the trigger condition is to trigger at 8 am every day; the data source is the order table in MySQL; the data target is the target order table in Hive; the conversion rule is to filter and delete records whose deletion identifier field is deleted; and the constraint condition is to complete the synchronization within a preset time window.

[0046] In some optional implementations, the first intelligent agent module 201 can also determine first knowledge query information based on the input information. The first knowledge query information may include scenario type, source system type, target system type, business object, triggering method, and constraints. For example, for the input information "synchronize the order table from MySQL to Hive every morning at 8 AM," the first knowledge query information may include "scheduled synchronization," "MySQL," "Hive," "order table," and "8 AM daily," etc.

[0047] The first intelligent agent module 201 obtains matching first knowledge information from the knowledge base unit based on the first knowledge query information. The first knowledge information can be similar historical cases, requirement parsing templates, synonym mappings, field naming rules, business terminology dictionaries, or scenario recognition rules, etc. For example, the knowledge base may contain cases of "timed data synchronization," which includes fields such as trigger time, source database, source table, target database, and target table.

[0048] Preferably, the first intelligent agent module 201 obtains the first knowledge information that matches the first knowledge query information from the knowledge base unit through semantic similarity matching.

[0049] If the matching degree between the first knowledge information and the input information meets a preset condition, the first intelligent agent module 201 can generate a first generation result based on the first knowledge information and the input information. The preset condition can be that the matching degree is greater than or equal to a preset threshold, or that the matching degree ranks highest and exceeds the minimum confidence threshold. For example, when the matching degree of similar cases is 0.82 and the preset threshold is 0.8, the structured requirement fields of that case can be reused and replaced with differences based on the current input information.

[0050] For example, if a user inputs "synchronize orders from PostgreSQL to the data warehouse daily," the knowledge base matches the case "case_001 (MySQL to Hive synchronization)," with a match score of 0.82, indicating a high similarity requirement. Therefore, replacements are made based on the differences: the source system is replaced from MySQL to PostgreSQL, and the target system is replaced from Hive to ClickHouse. Subsequent use of the matched case allows for reuse; only two components need to be replaced, while the remaining components and parameter configurations are retained. This significantly reduces workflow generation time and improves efficiency.

[0051] The second intelligent agent module 202 takes the first generation result output by the first intelligent agent module 201 as input information, determines the corresponding component planning information, and generates a second generation result including the component planning information.

[0052] The component planning information is used to characterize the workflow nodes corresponding to the structured description and the relationships between these workflow nodes. For example, the component planning information may include one or more of the following: a selected list of components, inter-component dependencies, and selection rationale. For instance, the selected list of components may include component node identifiers, component node types, component versions, and component sources; inter-component dependencies may include data transfer relationships, control dependencies, and exception handling relationships between component nodes; and the selection rationale may explain the basis for selecting the component, such as component matching with data source type, compatibility with the target system, meeting execution time windows, and validity verified by historical cases.

[0053] As an example, the second agent module 202 can be used as a component to plan the agent. When the first generated result indicates "timed triggering, reading from MySQL, filtering data, and writing to Hive", the second agent module 202 can determine the candidate nodes, including the timed triggering node, the database reading node, the data filtering node, and the data writing node. Specifically, it can include: node trigger_001 as the timed triggering node, node reader_001 as the database reading node, node filter_001 as the data filtering node, and node writer_001 as the data warehouse writing node; the dependency relationship is that trigger_001 points to reader_001, reader_001 points to filter_001, and filter_001 points to writer_001; the reason for selection is that the input information includes a daily 8 o'clock timed condition and the source system is MySQL and the target system is Hive.

[0054] In some optional implementations, the second agent module 202 can obtain second knowledge information from the knowledge base unit. This second knowledge information matches either the component planning information or the first generation result. The second knowledge information may include component templates, component applicable conditions, component dependencies, component mutual exclusion rules, successful component combination, and reasons for component selection.

[0055] The second intelligent agent module 202 generates a second generation result based on component planning information and second knowledge information, including one or more of the following: a list of selected components, dependencies between components, and reasons for component selection.

[0056] In some optional implementations, the second agent module 202 can also generate alternative component planning information. When multiple candidate paths (i.e., multiple component planning information) exist, the second agent module 202 can also sort the candidate paths based on information such as historical execution results, component reliability, resource version, and data volume in the knowledge base unit, and output the main path and alternative paths. If the task flow information corresponding to the subsequent main path fails, it can be switched to an alternative path.

[0057] The third intelligent agent module 203 is used to determine the parameter information corresponding to the component planning information based on the second generation result, and generate a third generation result including the parameter information, wherein the parameter information is used to characterize the execution parameters of the workflow node.

[0058] As an example, the third agent module 203 can be a parameter-filled agent used to fill specific parameters for components and coordinate parameter passing between components.

[0059] In an optional embodiment, the third intelligent agent module 203 can generate parameter information based on the first generation result and the second generation result. The first generation result provides information such as business semantics and user constraints, while the second generation result provides information such as component node types and dependencies. Based on this information, the third intelligent agent module 203 fills in parameters for each component node. For example, for a timed trigger node, it fills in a periodic expression and time zone; for a read node, it fills in a connection identifier, read table name, field list, batch size, and query conditions; for a filter node, it fills in a filter expression; and for a write node, it fills in the target table name, write mode, partitioning strategy, and failure retry parameters.

[0060] In an optional embodiment, the third intelligent agent module 203 can validate parameter information based on parameter constraint rules in the knowledge base unit. Parameter constraint rules may include, but are not limited to, one or more of the following combinations: parameter type, value range, mandatory fields, parameter dependencies, component input / output consistency, permission conditions, and environment adaptation rules. For example, the batch size must be a positive integer and within the range defined by the component template; when the write mode is overwrite, the target table permissions need to be confirmed; when the trigger condition is 8 AM daily, the timing expression should be consistent with the expression format of the selected execution engine.

[0061] In response to the parameter information passing verification, the third intelligent agent module 203 can further optimize the parameter information based on the knowledge base unit. Optimization can be based on successful parameters from historical cases, best practices in the rule base, default values ​​in the model base, and characteristics of the execution environment. For example, if historical cases show that a batch size of 5000 data entries performs well with the same amount of data, the batch size of the read node can be optimized to 5000 data entries; if the target system has an upper limit on concurrent writes, the parallelism can be limited to the allowable range.

[0062] The third agent module 203 generates a third generation result based on the second generation result and the optimized parameter information. The third generation result can be complete parameterized workflow information or an intermediate object used to generate workflow information. The third generation result can include, but is not limited to, one or more of the following combinations: node configuration, dependencies, parameter validation results, optimization descriptions, risk warnings, and items to be confirmed.

[0063] In one example, the third generated result includes: the execution parameters for trigger_001 are daily at 8 AM and the time zone is Beijing time; the execution parameters for reader_001 include connection identifier, source table orders, batch size of 5000 data entries, and read timeout; the execution parameters for filter_001 include the filter condition deleted=false; and the execution parameters for writer_001 include target table order_dw, write mode as append write, and number of retries.

[0064] Furthermore, to improve the collaboration efficiency between intelligent agent modules, the modules can transmit generated results through a target message protocol. The target message protocol can be used to standardize the transmission method of generated results, avoiding problems such as information omissions, ambiguities, and conflicts caused by simple message passing between different intelligent agent modules.

[0065] In one optional implementation, the message structure of the target message protocol may include at least a generation result and confirmation information. The generation result carries the structured or semi-structured data output by the current agent module, while the confirmation information indicates whether the generation result can be used by subsequent agent modules, whether manual confirmation is required, whether there are fields to be supplemented, or whether there are risk warnings, etc.

[0066] In an optional implementation, the target message protocol may further include information such as message identifier, source agent module identifier, target agent module identifier, generation task type, input summary, knowledge reference identifier, confidence level, verification status, error message, version number, and timestamp. The knowledge reference identifier is used to indicate which knowledge information in the knowledge base unit was referenced by the current generation result, facilitating subsequent tracing of the generation basis.

[0067] For example, in the message sent from the first intelligent agent module to the second intelligent agent module, the generated result can be a structured description, and the confirmation message can be "Requirements understood, start component planning"; in the message sent from the second intelligent agent module to the third intelligent agent module, the generated result can be component planning information, and the confirmation message can be "Component flow determined, please fill in parameters"; in the message sent from the third intelligent agent module, the generated result can be parameterized workflow information, and the confirmation message can be "Workflow ready, start execution".

[0068] The technical solution provided in this embodiment of the application can effectively solve the efficiency problem of intelligent agent module collaboration and avoid decision conflicts by clarifying the interaction methods and message formats between intelligent agent modules.

[0069] In an optional implementation, the agent unit may further include a fourth agent module, which is used to obtain execution feedback information during the execution of workflow information and trigger optimization processing based on the execution feedback information.

[0070] The fourth intelligent agent module can act as an execution monitoring agent. During the execution of workflow information, it obtains the execution status and performance indicators related to the workflow information and generates a fourth generation result. The fourth generation result can serve as execution feedback information and can be used as a basis for generating optimization strategies, updating knowledge base units, and re-invoking one or more other intelligent agent modules.

[0071] In addition, the fourth agent module can also obtain workflow-related information before and after the workflow information is executed, in order to optimize the workflow information.

[0072] The input to the fourth intelligent agent module may include the third generation result or workflow information generated based on the third generation result, and output real-time performance data accordingly, including component node execution status, data quality indicators, resource usage and overall health.

[0073] In an extended implementation, the input to the fourth intelligent agent module may further include the third generation result, workflow information formed based on the third generation result, workflow instance identifier, execution environment information, component node dependencies, component parameter configurations, and monitoring rules in the knowledge base unit, and accordingly determine the monitoring objects, monitoring dimensions, and collection frequency. For example, for a timed data synchronization workflow, the fourth intelligent agent module can determine which components need to be monitored most closely: the read speed of the read component nodes, the processing latency of the filter component nodes, the write success rate of the write component nodes, and the estimated completion time of the overall task.

[0074] The fourth intelligent agent module can obtain execution feedback information through callback interfaces, log collection interfaces, metric collection interfaces, message queues, database status tables, or runtime probes. Execution feedback information may include at least one of the following: node execution status, start time, end time, number of records processed, throughput, error rate, average latency, number of retries, resource utilization, exception information, data quality metrics, and health score.

[0075] In one optional implementation, while performing a monitoring task, the fourth agent module can also obtain knowledge information corresponding to the monitoring task from the knowledge base unit. For example, this knowledge information may include one or more of the following: monitoring metric definitions, anomaly detection rules, historical performance baselines, failure modes, optimization strategy templates, risk thresholds, and rollback rules. For instance, the knowledge base unit may contain a rule such as "when CPU utilization is consistently above 85% and the batch size is greater than 4000 data entries, prioritize reducing the batch size," and the fourth agent module uses this rule to determine whether the current execution feedback information meets the preset optimization conditions.

[0076] When the fourth agent module generates the fourth result, it can perform standardization, aggregation, and anomaly labeling on the collected raw execution data. For example, the fourth result may include: workflow instance identifier, running status of each workflow node, current bottleneck node, throughput trend, error rate trend, resource usage status, health score, whether preset optimization conditions are met, and the reason for triggering optimization. The fourth result can be used directly as execution feedback information, or as the basis for generating subsequent optimization strategies or as input to a large language model.

[0077] When the fourth agent module determines that the execution feedback information meets the preset optimization conditions, it can trigger optimization processing. Optimization processing may include inputting the fourth generated result into a large language model to obtain an optimization strategy, or it may include directly generating an optimization strategy based on a rule base, or simultaneously generating multiple candidate optimization strategies and ranking them by risk. For example, when the fourth agent module detects a continuous decrease in the throughput of the reading node, a continuous increase in CPU utilization, and no significant increase in the error rate, it can output the abnormal reason "suspected batch size too large or parallelism too high," and trigger an optimization strategy to adjust the batch size from 5000 data points to 3000 data points.

[0078] The fourth agent module can also feed back the fourth generation result to the first, second, or third agent modules to achieve partial regeneration. For example, when the execution feedback indicates that a component is unavailable or the interface version is incompatible, the feedback can be sent to the second agent module to reselect the component; when the execution feedback indicates that the parameter is out of bounds, permissions are insufficient, or the timeout is unreasonable, the feedback can be sent to the third agent module to regenerate or optimize the parameters; when the execution feedback indicates that the input requirements are ambiguous, causing the workflow goal to mismatch, the feedback can be sent to the first agent module to re-parse the input information or trigger manual confirmation.

[0079] In a specific example, the third intelligent agent module generates a third generation result including parameter configurations for the timed trigger node, database read node, data filter node, and data write node, and forms workflow information accordingly. After the workflow information begins execution, the fourth intelligent agent module continuously collects metrics such as recordsRead, recordsPerSecond, errorRate, cpuUsage, and memoryUsage for the reader_001 node. When the recordsPerSecond of reader_001 decreases from 5000 data points to 2000 data points, and cpuUsage remains above 90%, the fourth intelligent agent module determines that preset optimization conditions are met based on the historical performance baseline and anomaly rules in the knowledge base unit, and generates a fourth generation result including the bottleneck node reader_001, the anomaly type being resource bottleneck, and the suggested action being to reduce the batch size. Subsequently, optimized workflow information can be generated based on this fourth generation result, and the optimized parameters can be written into the knowledge base unit for reuse in subsequent similar workflows.

[0080] The fourth intelligent agent module provided in this application embodiment can simultaneously handle performance data acquisition and adjustment strategy triggering, enabling the generation of task flow information to have a complete execution feedback loop. Furthermore, it can dynamically adjust the workflow information generation process or the workflow information execution process based on feedback, achieving adaptive dynamic optimization. Adjustments can be made without waiting for the task flow information to finish execution or run again, thus effectively improving the efficiency and stability of task flow execution.

[0081] In one embodiment, the application further includes: obtaining execution feedback information during the execution of the workflow information; if, based on the knowledge base unit, it is determined that the execution feedback information meets preset optimization conditions, then generating an optimization strategy for adjusting the workflow information; and generating optimized workflow information based on the optimization strategy.

[0082] Execution feedback information can be obtained from the fourth intelligent agent module, log collection module, indicator collection module, or workflow engine callback interface, etc., and this application does not limit this. Execution feedback information is used to characterize the execution status of workflow information and / or the status of the execution device, etc., and may include, but is not limited to, one or more of the following combinations: node execution status, number of records processed, number of records processed per second, error rate, average latency, CPU utilization, memory utilization, number of retries, total execution time, estimated completion time, current bottleneck node, and health score.

[0083] Preset optimization conditions can be stored in knowledge base units or determined based on historical failure cases, anomaly detection models, or manual configuration. For example, preset optimization conditions may include, but are not limited to, one or more of the following combinations: error rate exceeding a threshold, continuous node failures, throughput below a certain percentage of the historical average, CPU utilization consistently exceeding a threshold, memory utilization consistently exceeding a threshold, estimated completion time exceeding a time window, and data quality indicators below a threshold.

[0084] When the execution feedback information meets the preset optimization conditions, an optimization strategy can be generated. The optimization strategy can include, but is not limited to, one or more of the following combinations: parameter tuning, path switching, resource expansion, and degraded execution. Parameter tuning can adjust batch size, parallelism, timeout, retries, cache size, etc.; path switching can switch continuously failing nodes to alternative components or alternative execution paths; resource expansion can increase parallel instances or allocate more computing resources; degraded execution can reduce the processing scope, skip non-critical steps, or delay the execution of low-priority nodes.

[0085] Considering the complex and diverse root causes of anomalies during workflow information execution (potentially parameter configuration issues, resource bottlenecks, data distribution changes, etc.), comprehensive analysis of multi-dimensional indicators (throughput, latency, error rate, CPU, memory, etc.) is required to pinpoint the problem. Furthermore, optimization strategies need to be generated for specific scenarios, as preset rules cannot cover all possibilities. Therefore, this application also provides an optional implementation method: in response to execution feedback information meeting preset optimization conditions, the execution feedback information is at least input into a large language model, and the optimization strategy for the workflow information output by the large language model is obtained. Through the multi-dimensional analysis and understanding capabilities of the large language model, optimization suggestions that conform to the business scenario can be effectively generated.

[0086] The information input into the large language model can include execution feedback, current workflow information, relevant knowledge rules, historical similar failure cases, and optional adjustment actions. To avoid the large language model outputting unexecutable strategies, constraints and optional strategy templates from the knowledge base unit can be input together, or the optimized strategy can be validated through the knowledge base unit after it is obtained.

[0087] For example, the initial parameters are a batch size of 5000 data entries and a parallelism of 2 data entries per second. During execution, feedback shows that the throughput drops from 5000 data entries per second to 2000 data entries per second, CPU utilization remains above 90%, and the knowledge base unit contains a rule that "when CPU utilization is above 85% and the batch size is greater than 4000 data entries, prioritize reducing the batch size." In this case, an optimization strategy can be generated to adjust the batch size from 5000 to 3000 data entries, and optimized workflow information can be generated. The optimized workflow information can preferably continue execution, or it can take effect in the next execution as needed.

[0088] Furthermore, after the optimization strategy is generated, it can be validated. Strategy validation can include rule verification, risk assessment, impact scope assessment, and rollback strategy generation. The optimization strategy is only updated in the currently executing workflow information if it meets both security and execution constraints.

[0089] This application also provides a knowledge base unit update method, which specifically includes: obtaining the execution result of workflow information, and storing some or all information related to the workflow information in the knowledge base unit based on the execution result of the workflow information.

[0090] If the workflow is executed successfully, one or more of the following can be written into the knowledge base as a success case: input information, first generated result, second generated result, third generated result, workflow information, execution metrics, and success factors. Success factors may include effective component combinations, appropriate parameter ranges, execution strategies that meet time windows, and configurations that avoid anomalies.

[0091] If a workflow fails, the failure mode can be logged and the knowledge base updated. Failure modes can include connection failure, insufficient permissions, out-of-bounds parameters, data type mismatch, insufficient resources, persistent node timeouts, and rate limiting on the target system. When updating the knowledge base, avoidance rules can be generated, such as limiting parallelism on the same target system or adjusting batch size for specific data volumes.

[0092] If the workflow optimization strategy takes effect successfully, the execution feedback information before and after optimization, the optimization strategy, and the optimization effect can be stored in the knowledge base unit. Subsequently, when the intelligent agent module performs parameter generation or optimization tasks, it can retrieve this optimization experience and use it as a basis for generation or as a constraint.

[0093] The knowledge base update mechanism provided in this application can transform historical execution experience into knowledge information that can be used in subsequent generation processes, making the generation results more stable and more in line with the actual execution environment as the number of times they are used increases.

[0094] To avoid redundant storage of case information, rule information, component templates, and optimization strategies, which can lead to knowledge redundancy and a gradual decrease in query efficiency, a deduplication mechanism can be introduced to optimize storage.

[0095] For example, when a new case is added to the database, we can first search the knowledge base to see if there is already a case with a high matching degree (such as a matching degree greater than 0.9). If so, the information of the new case will be overwritten and updated in the knowledge base. During the overwriting and updating process, the ID of the case being overwritten can be retained and other field information can be updated.

[0096] The following example will explain in detail how to generate workflow information based on knowledge base reuse cases.

[0097] Specifically, the user input requirement is: "Synchronize orders from PostgreSQL to the data warehouse daily." After analysis by the requirement analysis agent, a semantic similarity algorithm identifies a similar case in the knowledge base: case_001 (MySQL to Hive synchronization), with a matching degree of 0.82 (the pre-set matching degree threshold is 0.8, meeting the high matching criteria). Therefore, this case can be reused.

[0098] The reuse process is as follows: 1. Requirements Analysis: Agent identifies core differences: different source systems (MySQL → PostgreSQL) and different target systems (Hive → ClickHouse). 2. Component planning Agent adjustment Component selection: Use postgresql-reader instead of mysql-reader, use clickhouse-writer instead of hive-writer, and reuse case_001 for other components; 3. Parameter population Agent is adjusted based on best practices: batchSize is kept at 5000 data entries (case_001 verification is valid), and connectionTimeout is added (PostgreSQL feature).

[0099] After the above reuse process, the final workflow information generation time is greatly reduced, and the generation quality is comparable to that of non-reused cases.

[0100] The following is a detailed explanation of how to adaptively optimize workflow information using a specific example.

[0101] Specifically, the initial configuration in the task flow information is: batchSize=5000, parallelism=2. During the execution of the task flow information based on the initial configuration, the following data was monitored: Time T1: Throughput = 5000 / s, CPU = 60%, normal status; Time T2: Throughput = 3000 / s, CPU = 85%, performance degradation; Time T3: Throughput = 2000 / s, CPU = 95%, continued deterioration. Here, batchSize is the batch size, and parallelism is the degree of parallelism.

[0102] As the detection data deteriorates over time, it triggers an adaptive optimization process, including: 1. Root cause analysis: High CPU usage, suspected to be due to insufficient memory causing frequent garbage collection (GC); 2. Strategy generation: Option A is to reduce the batch size to 3000 data points (conservative), and Option B is to reduce the parallelism to 1 data point (aggressive). 3. Decision: Choose option A (lower risk); 4. Adjustment implemented: batchSize reduced from 5000 data entries to 3000 data entries; 5. After optimization: At time T4, throughput = 3500 / s, CPU = 75%, performance recovered; at time T5, throughput = 4000 / s, CPU = 70%, stable.

[0103] In addition, the statement "When CPU > 85% and batchSize > 4000, consider reducing batchSize" can be updated to the knowledge base to provide a reference for similar situations in the future.

[0104] Figure 3 This is an architectural diagram of the workflow processing apparatus provided in this application. Figure 4 A flowchart of the workflow processing apparatus provided in this application. Now combined with... Figure 3 and Figure 4 This paper describes one implementation scheme of the present application.

[0105] like Figure 3 As shown, the processing apparatus for the workflow may include an intelligent agent unit, a knowledge base unit, and an adaptive optimization engine. The intelligent agent unit may include a requirements analysis intelligent agent, a component planning intelligent agent, a parameter filling intelligent agent, and an execution monitoring intelligent agent. The knowledge base unit may include a case library, a rule library, and a model library. The adaptive optimization engine may include a performance monitoring module, an anomaly detection module, a policy generation module, and a policy verification module.

[0106] In this implementation scheme, the requirements analysis agent obtains user input information and parses it into a structured description. The structured description may include one or more of the following: triggering conditions, data source, data target, transformation rules, and constraints. When parsing the input information, the requirements analysis agent can query the knowledge base unit for similar historical cases, requirements parsing templates, and scene recognition rules, and generate requirements parsing results based on the retrieved knowledge information. This improves the accuracy and consistency of the requirements parsing results.

[0107] The component planning agent determines component planning information based on the structured description output by the requirements analysis agent. This information can include workflow nodes, dependencies between them, and rationale for component selection. The agent can query the knowledge base for component templates, adaptation rules, dependency rules, and historical component combination schemes, and determine the component combination that matches the structured description. For example, in a data synchronization scenario, the agent can determine component planning information including trigger nodes, data read nodes, data transformation nodes, and data write nodes.

[0108] The parameter-injection agent is used to determine parameter information based on component planning information and generate a parameterized workflow that includes this parameter information. The agent can query the knowledge base unit for parameter constraint rules, recommended parameter values, component parameter templates, and historical successful parameters, and perform parameter generation, validation, and optimization based on the retrieved knowledge. For example, the agent can determine timing expressions for trigger nodes, connection information, batch size, and read conditions for data read nodes, and target table, write mode, and retry parameters for data write nodes.

[0109] The execution monitoring agent is used to obtain execution feedback information during the execution of workflow information. Execution feedback information may include one or more of the following: node execution status, throughput, latency, error rate, resource utilization, number of retries, current bottleneck node, and health score. The execution monitoring agent can send the execution feedback information to the adaptive optimization engine or use it as the basis for updating knowledge base units.

[0110] like Figure 4 As shown, in one specific workflow, the user inputs natural language requirements into the requirements analysis agent. The requirements analysis agent queries historical cases in the knowledge base unit based on these requirements and receives matching cases returned by the knowledge base unit. The requirements analysis agent generates a requirements structure based on the input information and matching cases, and returns requirements confirmation information to the user, or directly sends the requirements structure to the component planning agent.

[0111] After receiving the requirement structure, the component planning agent queries the knowledge base unit for component templates and receives the component information returned by the knowledge base unit. Based on the requirement structure and component information, the component planning agent generates component planning information and returns component planning confirmation information to the user, or directly sends the component planning information to the parameter filling agent.

[0112] After receiving component planning information, the parameter-filling agent queries the knowledge base unit for parameter rules and optimization experience, and receives parameter constraints or suggestions returned by the knowledge base unit. Based on the component planning information, parameter constraints, and parameter suggestions, the parameter-filling agent generates a parameterized workflow and returns parameter configuration confirmation information to the user, or directly sends the parameterized workflow to the execution monitoring agent.

[0113] After receiving the parameterized workflow, the execution monitoring agent can control or trigger its execution and obtain execution feedback information during the process. The execution monitoring agent sends this feedback information to the adaptive optimization engine. The adaptive optimization engine determines whether the feedback information meets preset optimization conditions; if so, it generates an optimization strategy and returns it to the execution monitoring agent. The execution monitoring agent can then generate optimized workflow information based on the optimization strategy or call the parameter filling agent to regenerate parameter information. After the workflow execution is complete, the execution monitoring agent can send the execution results to the knowledge base unit, enabling the knowledge base unit to update cases, rules, or models.

[0114] pass Figure 3 and Figure 4 The implementation scheme shown allows multiple agents within the agent unit to collaboratively process input information in the order of requirement parsing, component planning, parameter filling, and execution monitoring; the knowledge base unit can provide corresponding knowledge information at different processing stages; and the adaptive optimization engine can trigger optimization processing based on execution feedback information. This forms a closed-loop processing procedure from input information acquisition, workflow generation, workflow execution, execution feedback, optimization adjustment to knowledge updating.

[0115] Figure 5 This is a schematic diagram illustrating the working principle of the adaptive optimization engine provided in this application. Now, combined with... Figure 5 This section explains the working principle of the adaptive optimization engine.

[0116] like Figure 5As shown, the adaptive optimization engine can include processing stages such as metric collection, anomaly detection, root cause analysis, strategy generation, and strategy verification. The metric collection stage is used to obtain execution feedback information during the workflow information execution process. Execution feedback information can include one or more of the following: throughput, latency, error rate, CPU utilization, and memory utilization. This execution feedback information can be obtained through execution monitoring agents, the workflow engine, log collection interfaces, metric collection interfaces, or runtime probes.

[0117] The anomaly detection stage determines whether the execution feedback information meets preset optimization conditions. This stage can include rule-based anomaly detection and model-based anomaly detection. Rule-based anomaly detection matches the execution feedback information against anomaly judgment rules in the knowledge base unit; model-based anomaly detection compares the execution feedback information with historical performance baselines or anomaly detection models. If the execution feedback information does not meet the preset optimization conditions, the current workflow can continue; if the execution feedback information meets the preset optimization conditions, the root cause analysis stage begins.

[0118] Root cause analysis is used to determine the causes of execution anomalies or performance degradation. Root cause analysis can be based on large language model analysis or on knowledge base query results. For example, when throughput decreases, CPU utilization increases, but the error rate does not increase significantly, the anomaly may be related to excessive batch size, excessive parallelism, or insufficient resources. When the error rate increases and the target node continues to retry, the anomaly may be related to connection anomalies, permission anomalies, target system rate limiting, or parameter mismatch.

[0119] The strategy generation stage generates at least one candidate optimization strategy based on the root cause analysis results. Candidate optimization strategies may include one or more of the following: parameter tuning, path switching, resource expansion, and degraded execution. Parameter tuning may include adjusting one or more of the following: batch size, parallelism, timeout, cache size, and retries; path switching may include switching the current component to a candidate component or switching the current execution path to a candidate execution path; resource expansion may include adding parallel instances or increasing resource quotas; degraded execution may include skipping non-critical nodes, reducing the processing scope, and delaying the execution of low-priority tasks.

[0120] The strategy verification stage is used to validate candidate optimization strategies before they take effect. Strategy verification can include executability checks, risk assessments, rule checks, and rollback checks. If a candidate optimization strategy passes verification, optimization workflow information can be generated based on that strategy, and this optimized workflow information can take effect in the current or subsequent execution processes. If a candidate optimization strategy fails verification, the process can return to the strategy generation stage to regenerate the optimization strategy, or the original workflow information can be maintained and execution can continue.

[0121] In a specific example, the initial parameters of the workflow information include a batch size of a first value and a parallelism of a second value. During execution, the execution feedback information obtained from the metric collection stage indicates that throughput is continuously decreasing and CPU utilization is continuously increasing. The anomaly detection stage, based on the historical performance baseline and anomaly judgment rules in the knowledge base unit, determines that the execution feedback information meets the preset optimization conditions. The root cause analysis stage determines that the cause of the anomaly is related to the batch size or parallelism. The strategy generation stage generates two candidate optimization strategies: reducing the batch size and reducing the parallelism. The strategy verification stage, based on the risk assessment results, determines that the risk of reducing the batch size is lower than the risk of reducing the parallelism, and therefore the optimization strategy of reducing the batch size can be prioritized. After implementing this optimization strategy, execution feedback information continues to be collected, and the optimization strategy knowledge in the knowledge base unit is updated according to the optimized execution effect.

[0122] pass Figure 5 The implementation scheme shown demonstrates that the adaptive optimization engine can perform anomaly detection, root cause analysis, strategy generation, and strategy verification based on execution feedback information during workflow information execution, enabling the workflow information to be dynamically adjusted according to the actual execution status. This improves the stability of the workflow execution process and allows effective strategies from the optimization process to be stored in a knowledge base unit for reuse in the generation or optimization of similar workflow information in the future.

[0123] This application also provides a workflow processing apparatus, comprising: an acquisition unit for acquiring input information; a knowledge base unit including knowledge information; and an intelligent agent unit for processing the input information based on the knowledge base unit to generate workflow information corresponding to the input information, wherein the workflow information can be executed to achieve a function corresponding to the input information; wherein the intelligent agent unit includes intelligent agent modules for executing different generation tasks, each intelligent agent module generating a generation result corresponding to its generation task, and at least one intelligent agent module using the generation result generated by its previous intelligent agent module as input to execute its corresponding generation task; wherein at least some of the intelligent agent modules, while executing their corresponding generation tasks, obtain knowledge information corresponding to the generation task from the knowledge base unit and execute the generation task based on the knowledge information.

[0124] This application also provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the workflow processing method described in any embodiment of this application.

[0125] This application also provides an electronic device including at least one processor and a memory connected to the processor, wherein the memory is used to store a computer program; the processor is used to execute the computer program to enable the electronic device to implement the workflow processing method described in any embodiment of this application.

[0126] This application also provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the workflow processing method described in any embodiment of this application.

[0127] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0128] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0129] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A workflow processing method, characterized in that, include: Obtain input information; The intelligent agent unit processes the input information based on the knowledge base unit to generate workflow information corresponding to the input information. The workflow information can be executed to achieve the function corresponding to the input information. The intelligent agent unit includes intelligent agent modules for performing different generation tasks. Each intelligent agent module generates a generation result corresponding to its generation task. At least one intelligent agent module takes the generation result generated by the previous intelligent agent module as the input result to execute its corresponding generation task. In this process, at least some of the intelligent agent modules, while performing their corresponding generation tasks, obtain knowledge information corresponding to the generation tasks from the knowledge base unit and execute the generation tasks based on the knowledge information.

2. The method according to claim 1, characterized in that, When the intelligent agent module is executing a corresponding generation task, it obtains knowledge information corresponding to the generation task from the knowledge base unit, including: Based on the type of the generated task corresponding to the intelligent agent module and the input result obtained by the intelligent agent module, knowledge query information is generated; Determine the knowledge information in the knowledge base unit that matches the knowledge query information; and The knowledge information is used as the basis for generating and / or constraining the intelligent agent module to execute the generation task.

3. The method according to claim 1, characterized in that, The intelligent agent unit includes: The first intelligent agent module is used to parse the input information, obtain a structured description corresponding to the input information, and generate a first generation result based on the structured description; The second intelligent agent module is used to determine the component planning information corresponding to the structured description based on the first generation result, and generate a second generation result including the component planning information. The component planning information is used to characterize the workflow nodes corresponding to the structured description and the association relationships between the workflow nodes. The third intelligent agent module is used to determine the parameter information corresponding to the component planning information based on the second generation result, and generate a third generation result including the parameter information, wherein the parameter information is used to characterize the execution parameters of the workflow node. The workflow information is generated based on the third generation result.

4. The method according to claim 3, characterized in that, The first intelligent agent module generates the first generation result including: The first intelligent agent module determines the first knowledge query information based on the input information; Obtain first knowledge information that matches the first knowledge query information from the knowledge base unit; and If the matching degree between the first knowledge information and the input information meets the precondition, then based on the first knowledge information and the input information, the first generation result is generated. The first generated result includes at least one of the following combinations: triggering condition, data source, data target, transformation rule, and constraint condition.

5. The method according to claim 3, characterized in that, The second intelligent agent module generates the second generation result, including: The second intelligent agent module generates component planning information based on the first generation result; Second knowledge information is obtained from the knowledge base unit, and the second knowledge information matches the component planning information or matches the first generation result; and Based on the component planning information and the second knowledge information, the second generation result is generated. The second generated result includes at least one of the following combinations: a list of selected components, dependencies between components, and reasons for component selection.

6. The method according to claim 3, characterized in that, The third intelligent agent module generates the third generation result, including: The third intelligent agent module generates parameter information based on the first generation result and the second generation result; The parameter information is verified based on the parameter constraint rules in the knowledge base unit; In response to the parameter information passing verification, the parameter information is optimized based on the knowledge base unit; and Based on the second generation result and the optimized parameter information, the third generation result is generated.

7. The method according to claim 1, characterized in that, The method further includes: Obtain execution feedback information during the execution of the workflow information; If, based on the knowledge base unit, it is determined that the execution feedback information meets preset optimization conditions, then an optimization strategy for adjusting the workflow information is generated, and Based on the optimization strategy, optimized workflow information is generated.

8. The method according to claim 7, characterized in that, The method further includes: In response to the execution feedback information satisfying preset optimization conditions, at least the execution feedback information is input into the large language model; and An optimization strategy for obtaining the workflow information output by the large language model.

9. The method according to claim 1, characterized in that, The method further includes: The intelligent agent modules transmit the generation result through a target message protocol. The message structure of the target message protocol includes at least: the generation result and confirmation information, and / or In response to the execution of the workflow information, the execution result of the workflow information is obtained; and based on the execution result of the workflow information, some or all information related to the workflow information is stored in the knowledge base unit.

10. A processing apparatus for a workflow, characterized in that, include: The acquisition unit is used to acquire input information; Knowledge base units include knowledge information; as well as An intelligent agent unit is used to process the input information based on the knowledge base unit to generate workflow information corresponding to the input information. The workflow information can be executed to achieve the function corresponding to the input information. The intelligent agent unit includes intelligent agent modules for performing different generation tasks. Each intelligent agent module generates a generation result corresponding to its generation task. At least one intelligent agent module takes the generation result generated by the previous intelligent agent module as the input result to execute its corresponding generation task. In this process, at least some of the intelligent agent modules, while performing their corresponding generation tasks, obtain knowledge information corresponding to the generation tasks from the knowledge base unit and execute the generation tasks based on the knowledge information.

Images