Workflow orchestration method and computing device cluster

Abstract

The embodiment of the application provides a workflow arrangement method and a computing device cluster, relates to the technical field of computing devices, and can reduce the technical threshold of a non-technical background user for configuring a workflow, and improves the efficiency and accuracy of workflow arrangement. The method is applied to a computing device cluster, wherein the computing device cluster comprises a plurality of packaged task capability (PTC) units, and the method comprises the following steps: matching business requirement description information of a user with metadata of each PTC unit to generate an initial workflow; in response to execution requirement description information input by the user for each target PTC unit in the initial workflow, processing the corresponding execution requirement description information by a business intelligent agent of each target PTC unit to generate business parameters of each target PTC unit; and determining technical parameters of each target PTC unit based on the business parameters of each target PTC unit and a preset mapping relationship between the business parameters and the technical parameters, so as to obtain a target workflow.

Description

Technical Field

[0001] This application relates to the field of computing device technology, and in particular to a workflow orchestration method and a computing device cluster. Background Technology

[0002] With the development of artificial intelligence technology, low-code AI application building platforms are gradually becoming important tools for empowering business personnel to achieve intelligent innovation. These platforms allow users to build AI workflows for specific business scenarios through a visual node orchestration approach, using drag-and-drop and configuration.

[0003] However, business personnel without a technical background may face a high cognitive burden during workflow orchestration due to the excessive exposure of underlying configuration details, resulting in low workflow orchestration efficiency and a high risk of errors. Summary of the Invention

[0004] This application provides a workflow orchestration method and a computing device cluster, which can reduce the technical threshold for non-technical users to configure workflows and improve the efficiency and accuracy of workflow orchestration.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, embodiments of this application provide a workflow orchestration method applied to a computing device cluster. The computing device cluster includes multiple packaged-task capabilities (PTC) units. A PTC unit is a pre-packaged executable task unit. Each PTC unit includes metadata and a business intelligence agent, wherein the metadata describes the execution capability characteristics of the corresponding PTC unit, and the business intelligence agent executes the business logic of the PTC unit.

[0006] The method includes: First, matching the user's business requirement description information with the metadata of each PTC unit in multiple PTC units to generate an initial workflow corresponding to the business requirement description information. The initial workflow includes the identifiers of multiple target PTC units that match the business requirement description information, as well as the execution order among the multiple target PTC units. Next, in response to the execution requirement description information input by the user for each target PTC unit in the initial workflow, the business intelligence agent of each target PTC unit processes the corresponding execution requirement description information to generate business parameters for each target PTC unit. The business parameters are used to describe the execution information of the corresponding target PTC unit from a business perspective. Finally, based on the business parameters of each target PTC unit and the preset mapping relationship between business parameters and technical parameters, the technical parameters of each target PTC unit are determined to obtain the target workflow corresponding to the business requirement description information.

[0007] The workflow orchestration method provided in this application introduces PTC units as the basic execution units of the workflow. Each PTC unit uses metadata to describe its execution capability characteristics, and a business agent parses execution requirements to generate business parameters and execute business logic, thus decoupling business semantics from underlying execution technology capabilities. Based on this, the PTC's metadata is matched with the user's business requirement description information to automatically generate an initial workflow. This allows business personnel to quickly obtain the workflow skeleton simply by describing their business intent, without needing to concern themselves with the underlying technical concepts of the PTC unit. Similarly, the business agent of each target PTC unit processes the user's input execution requirements and converts them into business parameters, transforming the PTC unit's technical configuration into a business requirement description. This effectively shields users from the cognitive and operational burden caused by excessive exposure of underlying configuration details, such as parameters, interfaces, and scheduling strategies. Finally, through the mapping relationship between business parameters and technical parameters, the business parameters are automatically converted into the underlying technical parameters required for the actual operation of the PTC unit, resulting in the target workflow. This method encapsulates the configuration process of underlying technical parameters within the PTC unit, enabling business personnel without a technical background to intuitively orchestrate workflows from a business perspective. This significantly lowers the technical threshold and improves the efficiency and accuracy of workflow orchestration.

[0008] In one possible implementation, the metadata of each PTC unit includes: functional type characteristics, and one or more of execution resource consumption characteristics and execution performance characteristics. The user's business requirement description information is matched with the metadata of each PTC unit among multiple PTC units to generate an initial workflow corresponding to the business requirement description information. This includes: First, identifying the business action sequence and business priority in the business requirement description information, where the business action sequence includes multiple business actions and their execution order. Second, semantically matching each business action with the functional type characteristics of each PTC unit to determine at least one candidate PTC unit corresponding to each business action. Then, among the candidate PTC units corresponding to each business action, matching the business priority with the execution resource consumption characteristics and / or execution performance characteristics of each candidate PTC unit to determine the target PTC unit corresponding to each business action. Finally, based on the identifiers and execution order of each target PTC unit, the initial workflow corresponding to the business requirement description information is generated.

[0009] In this way, based on the business action sequence and business priority of business needs, and combined with the functional type, execution resource consumption and execution performance of PTC units, hierarchical semantic matching and optimal selection are performed. Without the need for user intervention in the underlying resource and performance configuration, the target PTC unit that fits the business needs is automatically selected, further reducing the scheduling burden of business personnel and improving the rationality and execution reliability of the initial workflow.

[0010] In another possible implementation, the metadata of the PTC unit also includes data interface information, which describes the data structure of the corresponding PTC unit's input / output data. Based on the identifiers and execution order of each target PTC unit, an initial workflow corresponding to the business requirement description information is generated, including: determining the data mapping rules between adjacent target PTC units based on the execution order and the data interface information of each target PTC unit; and generating the initial workflow based on the identifiers, execution order, and data mapping rules of each target PTC unit.

[0011] Thus, by introducing the data interface information of the PTC unit, the data mapping and data flow format between target PTC units can be automatically derived according to the execution order, which helps to ensure the accurate flow of data between nodes, reduce the possibility of process failure due to data format mismatch or binding errors, and improve the robustness of workflow orchestration.

[0012] In another possible implementation, the business intelligence agent of each target PTC unit processes the corresponding execution requirement description information to generate business parameters for each target PTC unit. This includes: for any first target PTC unit, the business intelligence agent of the first target PTC unit performs semantic analysis on the execution requirement description information input by the user, extracting the execution elements of the first target PTC unit from the execution requirement description information. Execution elements at least include the execution target, execution constraints, and data source. Based on each execution element, business parameters are matched in a preset business parameter template to obtain the initial business parameters of the first target PTC unit. Each initial business parameter includes a parameter name and a recommended parameter value. The initial business parameters are displayed to the user. In response to the user's editing operation on the initial business parameters, the initial business parameters are updated to obtain the business parameters of the first target PTC unit.

[0013] In this way, the implementation process transforms the technical parameter configuration into a business language description. Users only need to adjust the business parameters according to their business knowledge to complete the underlying technical configuration of PTC, which helps to reduce the cognitive burden of single-node configuration while retaining the flexibility of manual intervention.

[0014] In another possible implementation, the recommended parameter values ​​are obtained in the following ways: extracted from the execution elements; or, based on the execution elements, a preset parameter value matching the execution elements is retrieved from a preset domain knowledge base and used as the recommended parameter value; or, based on the historical value records of the initial business parameters in the first target PTC unit, the historical frequency of each parameter value is counted, and the parameter value with the highest frequency is used as the recommended parameter value.

[0015] In this way, the recommended parameter values ​​integrate multi-source information such as execution element extraction, domain knowledge base retrieval, and historical frequency statistics, enabling parameter recommendations to adapt to current needs and reuse domain experience and historical data. This mechanism provides users with reasonable default parameter values, helping to reduce the number of trial and error attempts during the configuration process.

[0016] In another possible implementation, the method further includes: performing a test execution on the target workflow and obtaining the test execution results. Based on the business requirement description information, the expected execution result is determined, and the test execution result is compared with the expected execution result. The expected execution result includes at least one of the expected output format, expected output content, or expected execution metrics. If the comparison results are inconsistent, the target PTC unit in the target workflow is adjusted, and the target workflow is regenerated.

[0017] In this way, by testing and executing the target workflow and comparing the results with expectations, an automated feedback loop is formed. Users can use this to verify the effectiveness of the process and adjust deviations in a timely manner, which helps to improve the match between the final workflow and actual business needs and reduces the risk of rework after the process is put into production.

[0018] In another possible implementation, the computing device cluster also includes multiple historical workflow templates, each containing corresponding business metadata. The user's business requirement description information is matched with the metadata of each PTC unit to generate an initial workflow corresponding to the business requirement description information. This includes determining the semantic matching degree between the business metadata of each historical workflow template and the business requirement description information. If a historical workflow template with a semantic matching degree higher than a preset threshold exists, the historical workflow template with the highest semantic matching degree is used as the initial workflow corresponding to the business requirement description information; or, if no historical workflow template with a semantic matching degree higher than the preset threshold exists, the user's business requirement description information is matched with the metadata of each PTC unit to generate the initial workflow corresponding to the business requirement description information.

[0019] In this way, by matching historical workflow templates and reusing existing processes with high similarity, it helps to improve the orchestration efficiency of similar business scenarios, while semantic matching ensures the adaptability of templates to requirements.

[0020] In another possible implementation, the target workflow includes data mapping rules between each target PTC unit. The method further includes: responding to a user's business scheduling command for the target workflow, creating instantiated objects of each target PTC unit based on the identifiers and technical parameters of each target PTC unit in the target workflow. Based on the execution order, the instantiated objects of each target PTC unit are executed sequentially, and data conversion and transmission between the target PTC units are processed according to the data mapping rules to complete the business processing corresponding to the business requirement description information.

[0021] In this way, instantiated objects of each PTC unit are created based on the target workflow, and data transformation and transmission are automatically processed according to the execution order. This mechanism clarifies the automated execution path during workflow runtime, freeing users from the need to concern themselves with the underlying scheduling details of the workflow. This helps ensure the accurate execution of business logic and reduces the operational complexity after process deployment.

[0022] In another possible implementation, the PTC unit includes at least a base PTC unit and a domain PTC unit. The base PTC unit is an atomized PTC unit. The domain PTC unit is composed of the base PTC unit and / or other domain PTC units.

[0023] In this way, the reusability and flexibility of the smallest functional granularity can be guaranteed through the basic PTC unit, and complex capability units for specific business domains can be formed through multi-level combination, meeting the multi-level needs from fine-grained function calls to coarse-grained business process orchestration, and improving the efficiency and scalability of workflow construction.

[0024] Secondly, embodiments of this application provide a workflow orchestration apparatus for executing any of the workflow orchestration methods provided in the first aspect above.

[0025] Thirdly, embodiments of this application provide a computing device cluster, which includes at least one computing device, the computing device including a processor and a memory; the processor is coupled to the memory; the memory is used to store computer instructions, which are loaded and executed by the processor to enable the computing device cluster to implement the methods provided in the first aspect and its possible implementations described above.

[0026] Fourthly, embodiments of this application provide a computer-readable storage medium comprising: computer software instructions; and, when the computer software instructions are executed in a computing device, causing the computing device to implement the method provided by the first aspect and its possible implementations described above.

[0027] Fifthly, embodiments of this application provide a computer program product that, when run on a computing device, causes the computing device to execute the steps of the relevant method described in the first aspect above, so as to implement the method of the first aspect above.

[0028] The beneficial effects of the second to fifth aspects mentioned above can be referred to the corresponding description of the first aspect, and will not be repeated here. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the hardware structure of a computing device provided in an embodiment of this application; Figure 2 A schematic diagram of the software architecture of a computing device cluster provided in an embodiment of this application; Figure 3 A flowchart illustrating a workflow orchestration method provided in this application embodiment. Figure 1 ; Figure 4 A flowchart illustrating a workflow orchestration method provided in this application embodiment. Figure 2 ; Figure 5 A flowchart illustrating a workflow orchestration method provided in this application embodiment. Figure 3 ; Figure 6 A flowchart illustrating a workflow orchestration method provided in this application embodiment. Figure 3 ; Figure 7 A flowchart illustrating a method for generating PCT unit service parameters provided in an embodiment of this application; Figure 8 A flowchart illustrating a workflow orchestration method provided in this application embodiment. Figure 4 ; Figure 9 This is a flowchart illustrating a workflow execution method provided in an embodiment of this application. Detailed Implementation

[0030] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0031] In the description of this application, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can mean A or B. "And / or" in this application is merely a description of the relationship between the related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. A and B can be singular or plural.

[0032] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0033] Furthermore, to facilitate a clear description of the technical solutions in the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" are not necessarily different. Meanwhile, in the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is being used as an example, illustration, or description. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present related concepts in a concrete manner for ease of understanding.

[0034] The following is a brief introduction to the relevant terms used in the embodiments of this application.

[0035] 1. Intelligent Agent: Refers to a software entity capable of receiving input, understanding task intent, and autonomously executing corresponding logic. Technically, an intelligent agent may include components such as an AI model, a decision-making module, an execution module, and a knowledge base interface. In this application, the business intelligent agent built into each PTC unit is one type of intelligent agent.

[0036] 2. Intelligent Agent Workflow: (hereinafter referred to as workflow) refers to an execution process consisting of multiple intelligent agents or task units connected in a specific order to complete a specific business task. In this application, the workflow consists of multiple encapsulated task capability PTC units and their execution order and data mapping rules.

[0037] 3. PTC Unit: Refers to a pre-packaged, independently deployable, and reusable task execution unit. Each PTC unit contains metadata and a business intelligence agent. The metadata describes the execution capability characteristics of the corresponding PTC unit, while the business intelligence agent parses user execution requirements, generates business parameters, and executes business logic.

[0038] 4. Business Parameters: Data used to describe the execution information of the PTC unit from a business perspective. Business parameters express task execution requirements in business semantics, such as "response style = friendly" and "search scope = the last three years", without involving the details of the underlying technical implementation.

[0039] 5. Technical Parameters: Specific configuration parameters used to drive the execution of the underlying engine or model, directly related to the technical implementation. Examples include the temperature coefficient of a large language model and the top_k recall count for knowledge base retrieval.

[0040] The following section introduces the application scenarios of this application.

[0041] This application provides a workflow orchestration method that introduces a PTC unit as the basic execution unit for the workflow. It utilizes the metadata of the PTC unit to match the user's business requirements description at the business semantic level, automatically generating an initial workflow. Furthermore, it leverages the business intelligence agent built into the PTC unit to parse the user's input execution requirements into business parameters, and then converts these business parameters into underlying technical parameters through a preset mapping relationship. This decouples business semantics from underlying technology, enabling business personnel without a technical background to intuitively orchestrate workflows using business language. This effectively lowers the technical barrier to workflow orchestration and improves orchestration efficiency and accuracy.

[0042] The workflow orchestration method provided in this application can be applied to low-code AI application building platforms, enterprise-level intelligent business process management systems, and various industry digitalization scenarios that require business personnel to directly participate in the design of AI applications.

[0043] For example, in intelligent customer service scenarios, business operations personnel can quickly build complaint handling workflows using this method; in human resource management scenarios, HR professionals can use this method to construct resume screening and interview invitation processes; and in supply chain management scenarios, business personnel can flexibly configure inventory alerts and automatic replenishment strategies. This method is particularly suitable for frontline business scenarios where business needs change frequently and require rapid responses, enabling non-technical personnel in operations, product, and sales to independently build and adjust AI workflows, effectively shortening the digital delivery cycle of business processes.

[0044] The system architecture of this application will be described below.

[0045] The workflow orchestration method provided in this application embodiment can be applied to a computing device cluster, wherein the computing device cluster includes at least one computing device, which can be a blade server, a high-density server, a rack server, or a full rack server; functionally, the server can be a general-purpose server, a graphics processing unit (GPU) server, an artificial intelligence (AI) server, etc.

[0046] In an exemplary embodiment, such as Figure 1The diagram shown is a hardware structure diagram of a computing device provided in an embodiment of this application. The computing device includes: a memory 101, a transceiver 102, and at least one processor 103.

[0047] The transceiver 102 is used to interact with other devices to send and receive data. For example, in this embodiment, the transceiver 102 can be used to obtain the user's business requirement description information, or to send a workflow template to the user.

[0048] The memory 101 is used to store computer program code, which includes computer instructions. These computer instructions run in the aforementioned computing device to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include nonvolatile memory (NVM), such as at least one disk storage device, or a USB flash drive, portable hard drive, read-only memory, magnetic disk, or optical disk, etc.

[0049] Processor 103 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 103 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0050] The memory 101, transceiver 102, and processor 103 are communicatively connected. For example, the memory 101 and transceiver 102 can be connected to the processor 103 via a system bus to complete mutual communication. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0051] Alternatively, the memory 101 can be either standalone or integrated with the processor 103. When the memory 101 is set up independently, it is connected to the processor 103 via a system bus.

[0052] In an exemplary embodiment, such as Figure 2 The diagram shown is a software architecture diagram of a computing device cluster provided in an embodiment of this application, including: an interaction layer 201, a workflow layer 202, and a data layer 203. The overall process is divided into a configuration phase and an operation phase.

[0053] The data layer 203 is used to provide context information for workflow orchestration and execution, including: PTC library 2031, workflow library 2032, and knowledge base 2033.

[0054] The PTC library 2031 stores metadata for basic PTC units and domain PTC units.

[0055] The metadata for each PTC unit includes: identifier, metadata, business intelligence agent configuration, and domain knowledge base configuration. The metadata for a domain PTC unit also includes the internal execution order and data mapping rules. Optionally, the metadata may also include data interface information, default business parameter templates, and parameter mapping relationships.

[0056] Optionally, the metadata may also include: data interface information of the PTC unit, default business parameter templates, and mapping relationships between business parameters and technical parameters.

[0057] Workflow Library 2032 stores workflow metadata, which includes: the identifier of PTC cells in the workflow, execution order, and data mapping rules, etc.

[0058] Knowledge Base 2033 is used to provide domain knowledge data for the operation of business agents in the PTC unit.

[0059] For example, the content dimensions of domain knowledge data may include: facts (such as business rules, policy terms, and product information), principles (such as industry standards, technical specifications, and operating procedures), skills (such as script templates, response examples, and handling techniques), and interpersonal relationships (such as organizational structure, personnel responsibilities, and the preferences and permissions of communication partners).

[0060] Configuration phase: The interaction layer 201 receives orchestration instructions from users, including domain PTC unit orchestration and workflow orchestration instructions, which carry functional or business requirement description information.

[0061] Optionally, the orchestration instructions may also include a manual orchestration field, which indicates whether the current orchestration instructions are orchestrated automatically by the computing device cluster or manually by the user.

[0062] Workflow layer 202 is used to respond to user orchestration instructions and execute orchestration logic.

[0063] The workflow orchestration logic can be represented as follows: Based on the business requirement description information, the system matches the target PTC unit from the PTC library 2031 and generates an initial workflow; in response to the user's execution requirements input for each target PTC unit, it generates business parameters; it maps the business parameters to technical parameters; and it establishes domain knowledge data call relationships for the business agents of each target PTC unit to generate workflow metadata, which is then stored in the workflow library 2032.

[0064] Optionally, based on the execution requirements of each target PTC unit, access information for domain knowledge data can be configured for the business agent of each target PTC unit from the knowledge base 2033, serving as a reference for the business agent to parse execution requirements or generate business parameters.

[0065] The logic of the domain PTC unit orchestration can be expressed as follows: Based on the functional requirements description information, PTC units are matched from PTC library 2031 to determine the internal execution order and data mapping rules of each PTC unit. In response to the user's input of the execution requirements of each PTC unit, business parameters are generated and mapped to technical parameters to generate metadata of the new domain PTC unit, which is then stored in PTC library 2031.

[0066] It is understandable that the orchestration logic of domain PTC units is essentially the same as that of workflow orchestration. In practical applications, domain PTC units focus more on the encapsulation and reuse of atomic business capabilities, and their orchestration is usually used to build domain capability units that can be independently deployed and repeatedly invoked; while workflow orchestration is geared towards specific business needs scenarios, using domain PTC units as execution nodes, and focusing on the execution order, data dependencies, and data flow of the overall process.

[0067] Operation phase: Interaction layer 201 is used to receive user service scheduling instructions.

[0068] Workflow layer 202 is used to respond to the business scheduling instructions and execute the workflow. The workflow execution process includes workflow instantiation and workflow triggering execution.

[0069] The logic for workflow instantiation can be represented as follows: Based on business scheduling instructions, the target workflow is loaded from workflow library 2032; according to the unit identifier in the target workflow, the metadata of each target PTC unit is loaded from PTC library 2031; the domain knowledge data corresponding to the business intelligence agent of each target PTC unit is obtained from knowledge base 2033; instantiated objects of each target PTC unit are created; workflow context information is constructed according to the execution order and data mapping rules of the target workflow; the instantiated objects and context information are integrated to generate workflow instantiated objects. The instantiated objects of each target PTC unit are obtained.

[0070] Workflow trigger execution can be represented as: In response to workflow trigger events, the instantiated objects of each target PTC unit are called sequentially according to the execution order defined in the workflow instantiated object. When each unit is executed, input data is obtained from the workflow context according to the data mapping rules, and the execution result of each target PTC unit is updated to the context. After completing all target PTC calls, the final business processing result is output.

[0071] The embodiments of this application are described below.

[0072] In this embodiment, a plurality of PTC units are pre-built in the computing device cluster. Each PTC unit is a pre-packaged executable task unit, and each PTC unit includes: metadata of the PTC unit and a business intelligence agent of the PTC unit. The metadata is used to describe the execution capability characteristics of the corresponding PTC unit, and the business intelligence agent is used to execute the business logic of the PTC unit.

[0073] For example, metadata can take the form of, but is not limited to, text tags, coded identifiers, or hierarchical classification tags (e.g., domain → module → function).

[0074] The business intelligence agent of the PTC unit is a dedicated intelligence agent obtained by training with domain data or configuring business rules based on the business scenario and task type corresponding to the metadata.

[0075] In some embodiments, PTC units can be divided into basic PTC units and domain PTC units, depending on the complexity of the business scenario. A basic PTC unit is an atomic, single PTC unit. A domain PTC unit is composed of basic PTC units and / or other domain PTC units.

[0076] The basic PTC unit corresponds to an indivisible business operation, such as "calling a large language model to generate text" or "executing code snippets".

[0077] In an exemplary embodiment, as shown in Table 1, based on the business needs of a certain enterprise, six business types and several business operations are summarized. For each business operation under each object type, a corresponding basic PTC unit is constructed. Furthermore, {object-operation} is used as the metadata of the basic PTC unit.

[0078] Table 1

[0079] Domain-specific PTC units are composed of basic PTC units and / or other domain-specific PTC units, corresponding to complex business scenarios. For example, "customer complaint handling" can be obtained by combining "text preprocessing unit," "sentiment analysis unit," "knowledge base retrieval unit," and "response generation unit." In addition, "customer complaint handling" can also be used as a component of other domain-specific PTC units, such as combining with "satisfaction survey" to form "closed-loop customer complaint handling," realizing the nested reuse of business capabilities.

[0080] In this way, the reusability and flexibility of the smallest functional granularity can be guaranteed through the basic PTC unit, and complex capability units for specific business domains can be formed through multi-level combination, meeting the multi-level needs from fine-grained function calls to coarse-grained business process orchestration, and improving the efficiency and scalability of workflow construction.

[0081] like Figure 3 As shown, the workflow orchestration method provided in this application embodiment, when applied to the above-mentioned computing device cluster, may include the following steps S301-S303: S301. Match the user's business requirement description information with the metadata of each PTC unit in multiple PTC units to generate the initial workflow corresponding to the business requirement description information.

[0082] The initial workflow includes the identifiers of multiple target PTC units that match the business requirement description information, as well as the execution order among the multiple target PTC units.

[0083] Business requirement description information is used to characterize the business content that the user expects to complete. Metadata is used to describe the execution capability characteristics of the corresponding PTC unit, which can be understood as a comprehensive description of the business capabilities of the PTC unit.

[0084] The execution capability characteristics may include one or more of the following: function type characteristics, execution resource consumption characteristics, and execution performance characteristics. Specifically, the function type characteristics identify the business function category to which the PTC unit belongs, such as document creation, communication push, and data collection; the execution resource consumption characteristics describe the computing resources required for the PTC unit to execute, and may include, for example, the number of CPU cores, memory size, and disk capacity; and the execution performance characteristics describe the performance metrics of the PTC unit during execution, and may include, for example, task concurrency, processing latency, and data throughput.

[0085] The business requirement description information is matched with the metadata of each PTC unit. That is, at the business semantic level, the user's expected business content is matched with the business capabilities of the PTC unit, and the (target) PTC unit and workflow that can meet the user's business requirements are selected.

[0086] For example, the modalities of business requirement description information include, but are not limited to: text, images, audio, documents, and structured instructions. Furthermore, business requirement description information may also include combinations of multiple modalities.

[0087] In some embodiments, the computing device cluster can first convert non-textual business requirement description information into textual information, then perform semantic parsing on the textual description information to obtain business semantics, and match the business semantics with the metadata of each PTC unit to generate an initial workflow.

[0088] In one possible implementation, for the business requirement description information of the image modality, the computing device cluster uses optical character recognition technology to extract the text content in the image, or uses a multimodal understanding model to perform semantic parsing on the image content and generate the corresponding text description. After converting the image into text, the text is then semantically parsed to obtain the business semantics.

[0089] In another possible implementation, for the business requirement description information of the audio modality, the computing device cluster uses speech recognition technology to convert the audio into text, and then performs semantic parsing on the converted text to obtain the business semantics.

[0090] Based on the business semantics described in the business requirements, the computing device cluster further performs matching and sorting of PTC units: In one possible implementation, the computing device cluster determines the vector distance between the business semantics and the functional type features of each PTC unit, and filters out target PTC units whose similarity meets the threshold; then, by calculating the semantic association strength between any two PTC units, an initial directed graph with each PTC unit as a node is constructed.

[0091] For each pair of target PTC units {A, B}: if the semantic association strength from A to B is higher than a preset threshold and significantly higher than the reverse semantic association strength from B to A, then A is a prerequisite dependency of B. In this case, a directed edge from A to B is added to the directed graph. After traversing all unit pairs to form a directed graph, the execution order is obtained through topological sorting, generating the initial workflow.

[0092] In this way, all target PTC unit pairs in the initial directed graph are traversed. After the edge construction is completed, a directed graph structure of target PTC units is formed. The directed graph is then topologically sorted, and the sorting result is used to determine the execution order of each target PTC unit, thereby generating the initial workflow.

[0093] For example, if the business requirement description is "handling customer complaints", the target PTC units matched by the computing device cluster are: complaint classification PTC unit, knowledge retrieval PTC unit, response generation PTC unit, and work order creation PTC unit.

[0094] After calculating the semantic association strength between any two PTC units, the association strength of "complaint classification → knowledge retrieval" is much higher than that of the reverse "knowledge retrieval → complaint classification", "knowledge retrieval → response generation" is much higher than that of the reverse "response generation → knowledge retrieval", and "response generation → work order creation" is much higher than that of the reverse "work order creation → response generation". Based on this, corresponding directed edges are constructed. After sorting the topology of the resulting directed graph, the unit execution order is determined to be: complaint classification → knowledge retrieval → response generation → work order creation, and finally an initial workflow adapted to this business scenario is generated.

[0095] In another possible implementation, the above-mentioned business semantic transformation and matching process of business requirement description information can also be implemented by an intelligent agent. This intelligent agent is a pre-trained semantic understanding agent, which is trained through a large number of business requirement samples and corresponding PTC unit annotation data. It is used to directly map different forms of business requirement description information to the corresponding PTC unit combinations and their execution order.

[0096] For example, the agent can be a semantic understanding model pre-trained based on a deep neural network, a hybrid agent based on a combination of rules and learning, or a task-oriented agent that can call on an external knowledge base to assist in understanding.

[0097] In other embodiments, since the automatically generated initial workflow may not fully meet the user's personalized needs or the detailed requirements of a specific business scenario, the computing device cluster can also respond to the user's editing operation on the target PTC unit in the initial workflow after generating the initial workflow, so as to adjust, replace or add or delete the target PTC unit, and update the initial workflow based on the edited result.

[0098] In other embodiments, if the computing device cluster fails to find a PTC unit that matches the business requirement description information during the matching process, or if the business agent is unable to parse valid business parameters from the execution requirement description information, a prompt message can be returned to the user, suggesting that the user supplement or adjust the input content and try again.

[0099] S302. In response to the execution requirement description information input by the user for each target PTC unit in the initial workflow, the business intelligence agent of each target PTC unit processes the corresponding execution requirement description information and generates the business parameters of each target PTC unit.

[0100] Among them, the business parameters describe the execution information of the corresponding target PTC unit from a business perspective. The execution information can be understood as the business constraints used by the business agent in the PTC unit to characterize the execution of the PTC unit when executing business logic.

[0101] The execution requirement description information is used to characterize the execution effect, execution constraints, business processing rules, and personalized execution requirements that the user expects the target PTC unit to achieve. It refines the unit's execution standards to match the actual business scenario, preventing unit execution from deviating from user expectations. The format of the execution requirement description information is similar to that of the business requirement description information, and will not be elaborated further here.

[0102] Since the global business requirement description information only expresses the overall business content and cannot cover the differentiated execution details of each unit, it is necessary to provide targeted business guidance for each target PTC unit to make the generated business parameters more in line with the actual scenario, thereby making the technical parameters more accurate.

[0103] Meanwhile, the business intelligence agent corresponding to each target PTC unit has built-in professional knowledge data in the corresponding domain. It can specifically analyze the execution requirement description information input by the user, extract the execution elements related to its own function, and ensure that the generated business parameters conform to the business characteristics of the unit.

[0104] In one possible implementation, the business intelligence agent of each target PTC unit performs semantic parsing on the execution requirement description information input by the user, and identifies keywords that match the unit by combining the built-in domain knowledge data and the unit's own metadata; then, the identified keywords are converted into key-value pairs to generate business parameters corresponding to parameter names and parameter values.

[0105] For example, suppose a user inputs the following execution requirement description for creating a PTC unit for a document: "Use a template from a knowledge base to draft an email draft. The order data and customer data in the draft come from the data collection PTC." After semantic parsing this description, the business intelligence agent identifies keywords such as "use template," "draft email," "order data," "customer data," and "data collection PTC." These keywords are then transformed into key-value pairs to generate business parameters: Writing Project = Email Draft, Writing Requirements = Use Template, and Data Source = Data Collection PTC.

[0106] In another possible implementation, the metadata of the target PTC unit includes a default business parameter template. The computing device cluster can also match the identified keywords with the parameter items in the default business parameter template, and fill the corresponding parameter items with the successfully matched keywords to generate business parameters.

[0107] In some embodiments, the computing device cluster can accept multiple adjustments and optimizations by the user of the business parameters generated by the business agent. The business agent generates initial business parameters based on the user's execution requirement description information and displays them to the user. The user can modify or supplement the parameter values. The business agent regenerates the parameter configuration based on each adjustment by the user, and through multiple adjustments by the user, the business parameters gradually conform to the user's expectations.

[0108] S303. Based on the business parameters of each target PTC unit and the preset mapping relationship between business parameters and technical parameters, determine the technical parameters of each target PTC unit to obtain the target workflow corresponding to the business requirement description information.

[0109] Technical parameters are used to describe the operational configuration of the corresponding target PTC unit at the underlying execution level. Technical parameters are determined based on business parameters and their mapping relationships. This means that through the conversion from business semantics to technical semantics, the user-configured business rules are automatically translated into underlying parameters that the target PTC unit can directly execute, replacing the process of manually configuring underlying technical parameters.

[0110] For example, for PTC units based on large language models, technical parameters may include adjustable parameters such as the temperature coefficient of the large language model, the context window size, and the penalty coefficient; for PTC units for knowledge base retrieval, technical parameters may include retrieval recall, similarity threshold, index identifier, and query timeout; for PTC units for code execution, technical parameters may include runtime environment, memory limits, timeout, and dependency package configuration.

[0111] For example, following the example in S302, the execution requirements input by the user for creating a PTC unit for a document are parsed to generate the following parameters: Writing Project = Email Draft, Writing Requirements = Use Template, and Data Source = Data Acquisition PTC. Based on the mapping relationship, Writing Project = Email Draft corresponds to a temperature coefficient of 0.3 in the large language model, Writing Requirements = Use Template corresponds to the prompt word template identifier template_123, and Data Source = Data Acquisition PTC is mapped to the identifier of Data Acquisition PTC.

[0112] In one possible implementation, the computing device cluster pre-maintains a mapping table between business parameters and technical parameters. This table stores the correspondence between business parameter names and technical parameter names, as well as the conversion rules from business parameter values ​​to technical parameter values. For the business parameters of each target PTC unit, a matching lookup is performed in the mapping table to obtain the corresponding technical parameter name and conversion rule. Based on the current value of the business parameter, the corresponding technical parameter value is generated, thereby obtaining the complete technical parameter configuration of the target PTC unit.

[0113] In another possible implementation, the PTC unit has pre-defined mapping logic between business parameters and technical parameters. This logic is encapsulated within the PTC unit in the form of configuration files, rule scripts, or executable code. The computing device cluster inputs the business parameters to the corresponding PTC unit, which then generates and configures the technical parameters internally according to the pre-defined mapping logic and outputs directly executable technical parameters.

[0114] The workflow orchestration method provided in this application decouples and encapsulates business semantics and underlying technology by introducing a PTC unit. It automatically generates an initial workflow using metadata matching, allowing users to quickly build a process skeleton without needing to understand technical concepts. A business intelligence agent parses execution requirements to generate business parameters, transforming technical configurations into business descriptions and reducing cognitive burden. Based on a preset mapping relationship, the business parameters are automatically converted into technical parameters to obtain the target workflow. This method enables users without a technical background to intuitively orchestrate workflows from a business perspective, significantly lowering the technical threshold and improving orchestration efficiency and reliability.

[0115] The following examples illustrate the steps for generating the initial workflow.

[0116] Example 1 In some embodiments, in S301 above, the computing device cluster can perform more granular semantic analysis on the service requirement description information, decompose it into multiple service actions representing specific service operation steps, and then match a corresponding PTC unit for each service action based on the service actions. This process is as follows: Figure 4 As shown, S301 specifically includes steps S3011-S3014: For S302-S303, please refer to the above text. Figure 3 The details and related descriptions will not be elaborated here.

[0117] S3011. Identify the sequence of business actions and the business priority of the business requirement description information.

[0118] The business action sequence includes multiple business actions and their execution order. A business action is a single business operation unit that the user expects to complete. Examples include "querying an order," "sending an email," and "generating a report."

[0119] The execution order refers to the temporal dependencies between various business actions, such as action B can only be executed after action A is completed, or A and B can be executed in parallel.

[0120] Business priority is used to characterize the urgency or resource preference of a user for executing the overall business needs or a part of them.

[0121] In one possible implementation, the computing device cluster can perform semantic parsing based on natural language processing technology and a domain knowledge base. First, the business requirement description information is segmented, part-of-speech tagging is performed, and dependency parsing is conducted to extract the core verbs and their corresponding objects, combining them into business actions in the form of "verb + object." Then, the extracted business actions are compared with standard business actions stored in the domain knowledge base, and synonymous expressions are normalized to obtain standardized business actions. Finally, the execution order of each business action is determined based on logical connectors in the text.

[0122] At the same time, business priorities are identified based on degree adverbs in the descriptive information or the type of business object.

[0123] In another possible implementation, the computing device can also combine user identity characteristics (such as job level and department), historical behavioral data (such as priority preferences and execution frequency of similar past needs) or current business scenario context information (such as business time period and system load status) to more accurately determine business priorities.

[0124] For example, business priorities can be discrete values ​​(such as numerical levels: high, medium, low, or 1-5 levels) or continuous values ​​(such as priority scores between 0 and 1) to characterize the relative urgency of business needs.

[0125] S3012. Semantically match each business action with the functional type features of each PTC unit to determine at least one candidate PTC unit corresponding to each business action.

[0126] Among them, the candidate PTC unit is a PTC unit that can complete the corresponding business action, that is, the functional type feature of the PTC unit matches the semantics of the business action and has the potential to execute the business action.

[0127] In one possible implementation, the computing device cluster converts each business action into a semantic vector and calculates its similarity with the functional type features of the metadata of each PTC unit to determine the semantic similarity between each PTC unit and each business action. For each business action, PTC units whose similarity meets the similarity threshold are identified as candidate PTC units corresponding to that business action.

[0128] In another possible implementation, the computing device cluster adopts a knowledge graph-based matching method, which pre-constructs a knowledge graph that maps business actions to PTC unit functional type features. Candidate PTC units are determined by querying the PTC units corresponding to business actions in the knowledge graph.

[0129] In other embodiments, if a certain service action fails to match any PTC unit, the computing device cluster may send a prompt message to the user, suggesting that the user supplement the specific description of the service action or manually select the corresponding PTC unit; or, mark the service action as a node to be processed, to be configured by the user later.

[0130] S3013. In the candidate PTC units corresponding to each business action, the target PTC unit corresponding to each business action is determined by matching the business priority with the execution resource consumption characteristics and / or execution performance characteristics of each candidate PTC unit.

[0131] By screening candidate PTC units for each business action, it can be ensured that the resource consumption and performance indicators of the target PTC unit match the business priority. Specifically, high-priority business actions can be matched with PTC units with sufficient performance and high resource guarantee, ensuring that the task is executed quickly and stably; low-priority business actions are matched with PTC units with lower resource consumption, reducing the problem of execution resource mismatch and achieving an optimal match between business needs and system resources.

[0132] In one possible implementation, the computing device pre-stores a mapping relationship (such as a mapping table) between service priorities and resource consumption levels and performance levels. This mapping relationship defines the threshold range of resource consumption indicators (such as the number of CPU cores, memory size, disk I / O, etc.) and performance indicators (such as maximum latency, minimum concurrency, throughput, etc.) corresponding to different service priority levels (such as high, medium, and low).

[0133] Based on this mapping relationship, the following matching steps are performed in the candidate PTC unit corresponding to each business action: 1. Query the mapping relationship table based on business priority to determine the resource consumption indicators and performance indicators required by user business needs.

[0134] 2. Match the resource consumption indicators with the execution resource consumption characteristics of each candidate PTC unit to obtain the first score.

[0135] 3. Match the performance metrics with the execution performance characteristics of each candidate PTC unit to obtain the second score.

[0136] 4. Combine the first score and the second score to obtain the matching score, and determine the candidate PTC unit with the highest matching score as the target PTC unit for the corresponding business action.

[0137] In another possible implementation, the computing device cluster takes service priority as an input parameter and dynamically generates matching constraints on execution resource consumption characteristics and / or execution performance characteristics through a preset mapping function or model.

[0138] Specifically, based on the business priority value (such as 0~1 or 0.1~1.0), the corresponding resource and performance thresholds are calculated in real time through a mapping function.

[0139] For example, maximum allowable latency = baseline latency × (1 - priority), minimum throughput = baseline throughput × priority, and minimum number of CPU cores = baseline number of cores × priority, etc.

[0140] Then, these dynamically generated thresholds are used as constraints to compare the execution capability characteristics of each candidate PTC unit one by one to determine whether they meet the constraints (such as actual latency ≤ maximum allowable latency, actual throughput ≥ minimum throughput); the candidate PTC units that meet the constraints are determined as the target PTC units.

[0141] In this way, the method replaces static table lookup mapping with functional mapping, which can realize a continuous and smooth conversion from business priority to resource performance constraints, avoid the threshold jump problem that may be caused by discrete mapping, and make the matching of priority and PTC unit more flexible and accurate.

[0142] S3014. Based on the identifiers and execution order of each target PTC unit, generate the initial workflow corresponding to the business requirement description information.

[0143] The initial workflow uses the target PTC unit as a node and the execution order as the directed connection relationship between the nodes, forming an ordered task execution chain.

[0144] In some embodiments, the initial workflow may take the form of, but is not limited to, a JSON-formatted list of nodes and edges, an XML-formatted process definition file, or workflow records in a database table. The initial workflow shall at least include: the identifier of each target PTC unit and the execution order of each target PTC unit.

[0145] In one possible implementation, the computing device cluster constructs a directed graph with business actions as nodes according to the execution order identified by S3011, and then replaces each business action node in the directed graph with the target PTC unit matched by S3012 to obtain a directed graph with the target PTC unit as nodes. The topology sorting result of this directed graph is the initial workflow.

[0146] In this way, by semantically analyzing the business requirement description information and breaking it down into multiple business actions, and by matching the corresponding PTC unit to each business action based on the metadata of the PTC unit, a workflow can be constructed. This improves the alignment between the business semantics of the initial workflow and the user's business requirements, thereby enhancing the accuracy of the initial workflow construction.

[0147] Example 2 In some embodiments, the metadata of a PTC unit may also include data interface information, wherein the data interface information is used to describe the data structure of the input / output data of the corresponding PTC unit.

[0148] For example, the input data structure includes: the format of the input data (e.g., the input data is structured data, or the input data is a document file), the required fields / contents in the input data, the field types, the field value ranges, and the source constraints of the input data. The output data structure includes: the format of the output data, the fields / contents contained in the output data, the field types, and the organization of the output data.

[0149] For example, the input data structure of the Complaint Classification (PTC) unit is text format, with a required field being the complaint description; the output data structure is structured data, with output fields including complaint type, priority, and a list of tags. The input data structure of the Knowledge Retrieval (PTC) unit is structured data, with a required field being the complaint type; the output data structure is document data, with output fields including knowledge title, matching content, and source document.

[0150] In practical applications, the input requirements, output content, and dependent data sources of different functional nodes often differ, and subsequent functional nodes require the output of previous nodes as input to execute correctly. For complex workflows, the number of functional nodes is large and the data dependencies are complex, making manual sorting and configuration of data mappings costly and inefficient.

[0151] Based on this, in the embodiments of this application, the computing device cluster can automatically complete the data adaptation and mapping configuration within the process by relying on the data interface information of each PTC unit, without the need for manual sorting and docking. This process is as follows: Figure 5 As shown, based on Embodiment 1, S3014 may specifically include S30141-S30142: S3011. Perform semantic parsing on the business requirement description information to determine the multiple business actions corresponding to the business requirement description information and the relationships between the business actions.

[0152] S3012. Match each business action with the metadata of each PTC unit to determine the target PTC unit corresponding to each business action.

[0153] S3013. Based on the relationships between various business actions, determine the execution order between each target PTC unit.

[0154] For detailed descriptions of S3011-S3013 above, please refer to... Figure 4 The details and related descriptions will not be elaborated here.

[0155] S30141. Based on the execution order and the data interface information of each target PTC unit, determine the data mapping rules between adjacent PTC units.

[0156] In this context, adjacent PTC units refer to target PTC units that have a direct sequential relationship in the execution order. The data mapping rules describe the field correspondence, type conversion methods, and data format conversion rules between the output data of the preceding PTC unit and the input data of the subsequent PTC unit.

[0157] In one possible implementation, for each pair of adjacent PTC units, the output fields of the preceding PTC unit are semantically matched with the input PTC fields of the subsequent unit: direct mapping is established for fields with the same name or consistent business meaning; supplementary conversion rules are added for mismatched field types; supplementary standardization processing is added for inconsistent data formats; for fields required by the subsequent unit but not provided by the preceding unit, they are marked as to be supplemented or default values ​​are used. Once the mapping relationships between all adjacent units are determined, a complete data mapping configuration is formed.

[0158] For example, suppose that in the adjacent PTC unit "Complaint Classification PTC" → "Knowledge Retrieval PTC", the output fields of the "Complaint Classification Unit" include: classification label (string type) and confidence score (floating-point type); the input fields of the "Knowledge Retrieval Unit" include: query keywords (string type) and search scope (string type). Through semantic matching, the "classification label" is mapped to the "query keywords" to establish a direct mapping; while the "search scope" field has no corresponding field in the output of the preceding unit, so it is marked as to be supplemented or the default value "all" is used.

[0159] S30142. Generate an initial workflow based on the identifier, execution order, and data mapping rules of each target PTC unit.

[0160] Unlike the initial workflow generated by S3014, S30142 uses the target PTC unit as a node, the execution order as the directed connection relationship between the nodes, and adds data mapping rules as edge attributes on the node connection lines to form a complete task execution link containing data flow information.

[0161] In one possible implementation, the computing device cluster constructs a directed graph with business actions as nodes based on the business action relationships identified in S3011. Then, each business action node in the directed graph is replaced with the target PTC unit matched in S402, resulting in a directed graph with the target PTC unit as nodes. Based on this, the data mapping rules determined in S30141 are used as additional attributes of the corresponding edges in the directed graph; that is, each directed edge includes field mapping rules and transformation configurations from the source unit to the target unit. This directed graph with data mapping attributes constitutes the final generated initial workflow.

[0162] Thus, by introducing the data interface information of the PTC unit, the data mapping and data flow format between target PTC units can be automatically derived according to the execution order, which helps to ensure the accurate flow of data between nodes, reduce the possibility of process failure due to data format mismatch or binding errors, and improve the robustness of workflow orchestration.

[0163] Example 3 In some embodiments, there are Figure 2 As can be seen from its description, the workflow library in the computing device cluster can store templates of historical workflows. Each historical workflow template includes corresponding business metadata, which identifies the business type of the corresponding historical workflow template.

[0164] For example, business metadata can be obtained by transforming the business semantics of the user business requirement description information corresponding to historical workflows.

[0165] In this scenario, upon receiving the user's business requirement description, the computing cluster can first match the user's business requirement description with the business metadata of each historical workflow template. If a match is successful, the historical workflow template can be directly reused. If not, the workflow orchestration process then begins. This process is as follows: Figure 6 As shown, S301 specifically includes S301a-S301c: S301a. Determine the semantic matching degree between the business metadata and business requirement description information of each historical workflow template.

[0166] In one possible implementation, the computing device cluster converts the business requirement description information into semantic vectors, and at the same time converts the business metadata of each historical workflow template into semantic vectors. It calculates the cosine similarity or Euclidean distance between the business requirement vector and each template label vector, and uses the similarity score as the semantic matching degree.

[0167] In another possible implementation, the computing device cluster performs semantic parsing on the business requirement description information, extracts the business intent keywords, and then matches each business intent keyword with the business metadata of each historical workflow template. The number of successfully matched keywords is counted, and the matching score is determined as the proportion of the number of successfully matched keywords to the total number of keywords.

[0168] Based on this, if there is a historical workflow template with a semantic matching degree higher than the preset threshold, S301b is executed.

[0169] S301b: If there is a historical workflow template with a semantic matching degree higher than a preset threshold, the historical workflow template with the highest semantic matching degree shall be used as the initial workflow corresponding to the business requirement description information.

[0170] When a historical template that closely matches the current business needs exists, directly reusing the template can significantly reduce the workload of process construction. Furthermore, reusing a validated template helps ensure the rationality and stability of the workflow.

[0171] In one possible implementation, the computing device cluster compares all semantic matching scores calculated by S301a with a preset threshold, and filters out a set of candidate templates with matching scores higher than the threshold. If the candidate set is not empty, the template with the highest matching score is selected, and the PTC unit combination, execution order, and data mapping rules contained in the template are directly used as the initial workflow corresponding to the current business requirement.

[0172] In other embodiments, if there are multiple historical templates with semantic matching degrees all higher than the threshold and similar scores, the computing device cluster can push these candidate templates to the user, who can then manually select one as the initial workflow based on the actual scenario; or, multiple templates can be merged, and the common PTC units and execution order in each template can be taken as the basic skeleton of the initial workflow.

[0173] In other embodiments, when there are some differences between the matched historical template and the current business requirements, the computing device cluster can make adaptive adjustments based on the template: retain the part of the template that matches the current requirements, and re-execute the PTC unit matching process in S301 for the part that does not match, to generate a hybrid initial workflow.

[0174] Alternatively, if there is no historical workflow template with a semantic matching degree higher than the preset threshold, execute S301c.

[0175] S301c: In the absence of a historical workflow template with a semantic matching degree higher than a preset threshold, the user's business requirement description information is matched with the metadata of each PTC unit in multiple PTC units to generate the initial workflow corresponding to the business requirement description information.

[0176] This process can be referred to in S301 above, and will not be repeated here.

[0177] Then, after executing S301b or S301c, the computing device cluster continues to execute S302 and S303.

[0178] In this way, by matching historical workflow templates and reusing existing processes with high similarity, it helps to improve the orchestration efficiency of similar business scenarios, while semantic matching ensures the adaptability of templates to requirements.

[0179] The following section provides a detailed description of how the business intelligence agent processes and executes the requirement description information to generate business parameters.

[0180] In some embodiments, the computing device cluster can use a business intelligence agent to perform deep analysis of the execution requirement description information input by the user, identify key elements affecting the execution of the PTC unit, and transform these elements into structured business parameters based on a preset business parameter template. Simultaneously, it allows the user to adjust and confirm the generated parameters to ensure the accurate communication of business intent. For example... Figure 7 As shown, taking any first target PTC unit as an example, S302 specifically includes steps S3021-S3024: S3021. For any first target PTC unit, the business intelligence agent of the first target PTC unit performs semantic analysis on the execution requirement description information input by the user, and extracts the execution elements of the first target PTC unit from the execution requirement description information.

[0181] Among them, the execution elements include at least the execution objective, execution constraints, and data sources.

[0182] The execution target describes the result that the user expects the output of the first target PTC unit. The execution constraints describe the limitations or preferences during the execution process. The data source specifies the provider or reference location of the data required for the execution.

[0183] The semantic analysis process can be referred to the implementation method in S3011 above, and will not be repeated here.

[0184] For example, for the "Document Generation PTC Unit," the user-input execution requirement description is: "Based on Q3 quarterly data from the sales database, generate a business-style PPT document containing charts, highlighting the year-on-year growth rate." The execution elements extracted by the business intelligence agent through semantic analysis include: the execution goal is "generate a PPT document," the data source is "Q3 quarterly data from the sales database," and the execution constraints include "business style," "includes charts," and "highlights the year-on-year growth rate."

[0185] S3022. Based on each execution element, perform business parameter matching in the preset business parameter template to obtain the initial business parameters of the first target PTC unit.

[0186] Each initial business parameter includes: a parameter name and a recommended parameter value. The business parameter template predefines the types of business parameters that may be involved in this PTC unit, their value ranges, and the association rules between parameters.

[0187] The business intelligence agent performs semantic matching between the execution elements extracted by S3021 and the parameter items in the template: the execution target is mapped to the corresponding parameter name, the execution constraint is mapped to the parameter value or value range, and the data source is mapped to the data reference configuration, thereby generating a set of initial business parameters.

[0188] In one possible implementation, the business intelligence agent traverses each execution element extracted in S3021 and performs keyword matching between each element and parameter items in a preset business parameter template. If an element matches the trigger keyword of a parameter item, the element is mapped to that parameter item, and the parameter value is determined based on the element content to generate initial business parameters containing the parameter name and recommended value.

[0189] In some embodiments, because users have varying degrees of clarity regarding parameter values ​​in different scenarios, and some parameter values ​​reflect domain best practices or historical usage preferences, it is necessary to introduce multi-source information to supplement the recommendation basis. Therefore, in addition to obtaining the recommended parameter values ​​directly from the execution elements, they can also be determined through other methods: The domain knowledge base pre-stores recommendation parameter configurations for common business scenarios, which can make up for the lack of user input information; historical execution records reflect the user's habitual choices in similar scenarios, making the recommendation results more in line with the user's personalized preferences.

[0190] In one possible implementation, preset parameter values ​​that match the execution elements are retrieved from a preset domain knowledge base and used as recommended parameter values.

[0191] In another possible implementation, when the PTC unit has historical execution records, based on the historical value records of the initial business parameters in the first target PTC unit, the historical frequency of each parameter value is counted, and the parameter value with the highest frequency is used as the recommended parameter value.

[0192] For example, regarding the "Document Style" parameter in the "Document Generation PTC Unit," the distribution of values ​​in the past 100 execution records of this unit was analyzed: "Business" style appeared 65 times, "Minimalist" style appeared 25 times, and "Lively" style appeared 10 times. When the user does not explicitly specify a document style for this instance, the most frequently used "Business" value is used as the recommended parameter value.

[0193] In this way, the recommended parameter values ​​integrate multi-source information such as execution element extraction, domain knowledge base retrieval, and historical frequency statistics, enabling parameter recommendations to adapt to current needs and reuse domain experience and historical data. This mechanism provides users with reasonable default parameter values, helping to reduce the number of trial and error attempts during the configuration process.

[0194] S3023. Display initial business parameters to the user.

[0195] After generating initial business parameters, the business intelligence agent displays them to the user for confirmation or adjustment. On the one hand, since natural language parsing may contain ambiguities or omissions, displaying the initial business parameters allows users to verify their accuracy; on the other hand, users may have more refined and personalized requirements for certain parameters, and displaying them provides an adjustment entry point.

[0196] In one possible implementation, the computing device cluster presents the initial business parameters in a visual form in the interaction layer. For example, it can use a parameter configuration panel to display the name, recommended value and optional value range of each parameter in groups according to parameter category, highlight key parameters, and provide controls such as edit boxes, drop-down menus or switch buttons for users to modify.

[0197] S3024. In response to the user's editing operation on the initial service parameters, update the initial service parameters to obtain the service parameters of the first target PTC unit.

[0198] For example, a user's editing operations on initial business parameters may include: changing the business parameter name, changing the parameter value, adding missing parameters, or deleting redundant parameters.

[0199] For example, the initial business parameters generated for the "Document Generation PTC Unit" are "Output Format = PPT", "Document Style = Business", "Data Source = Sales Database", "Chart Requirements = Included", and "Key Indicators = Year-on-Year Growth Rate". After the user reviews the document, they change the "Document Style" to "Simple" and add "Page Limit = No more than 10 pages". The computing device cluster then updates the parameter configuration to obtain the final business parameters.

[0200] In this way, the implementation process transforms the technical parameter configuration into a business language description. Users only need to adjust the business parameters according to their business knowledge to complete the underlying technical configuration of PTC, which helps to reduce the cognitive burden of single-node configuration while retaining the flexibility of manual intervention.

[0201] The following section provides a detailed description of the testing, verification, and adjustment process for the target workflow.

[0202] In some embodiments, to verify whether the generated target workflow performs as expected by the user in actual execution, the computing device cluster can perform a test execution of the target workflow after S303 and iteratively optimize it based on the test results. This process is as follows: Figure 8 As shown, after S303, the method further includes steps S304-S306: S304. Perform test execution on the target workflow and obtain the test execution results.

[0203] The test execution involves running the execution process once or multiple times in an isolated test environment or sandbox environment according to the configuration of the target workflow to verify the actual operation effect of the workflow.

[0204] For example, the test execution results include: the final output of the target workflow, snapshots of the input / output data of each target PTC unit, workflow execution time, runtime exception information, resource consumption (such as memory usage, number of API calls), and data transfer records between units.

[0205] In one possible implementation, the computing device cluster sequentially calls each target PTC unit according to the execution order defined in the target workflow. Input data is passed according to data mapping rules. Input data snapshots and output data are recorded before and after the execution of each unit. At the same time, the execution time, logs, exceptions, and resource consumption indicators are collected. After all units have been executed, the final output results and the execution records of each unit are summarized, and a complete test execution result report is generated in time series.

[0206] In other embodiments, for workflows containing parallel execution units, parallel branches need to be started synchronously during test execution, and the results need to be summarized after all branches have been executed; for workflows with loops or conditional branches, test execution needs to cover different branch paths to verify the correctness of the workflow under various conditions.

[0207] S305. Determine the expected execution result based on the business requirement description information, and compare the test execution result with the expected execution result.

[0208] The expected execution result includes at least one of the following: expected output format, expected output content, or expected execution metrics. The expected output format describes the data format that the execution result should have; the expected output content describes the key content that the execution result should include; and the expected execution metrics describe the performance requirements that the workflow should meet during execution, such as a response time of no more than 5 seconds and resource consumption below a specific threshold.

[0209] In one possible implementation, the computing device cluster performs semantic parsing on the business requirement description information to identify the user's expected execution result for the target workflow.

[0210] For example, from "Generate a PDF report containing Q3 sales data", the expected output format is PDF and the expected output content should include Q3 sales data; from "Handle complaints and respond quickly", the expected execution indicator is a response time of no more than 3 seconds. If the expected requirements are not explicitly mentioned in the business requirement description information, the corresponding expected execution result standard is obtained from the preset default expectation library based on the function type characteristics in the PTC unit execution capability characteristics.

[0211] In one possible implementation, the computing device cluster extracts the format information, content fields and values, actual execution time and resource consumption of the final output data from the test execution results. Then, it matches and verifies these extracted items with the corresponding items in the expected execution results: the output format is compared to determine whether they are completely consistent, the output content is compared to determine whether the necessary fields exist and whether their values ​​are within the expected range, and the execution indicators are compared to determine whether the actual time and consumption are within the preset threshold.

[0212] In addition, for projects with deviations, record specific discrepancies (such as missing expected fields, values ​​outside the range, or exceeding time limits); for scenarios where the output content is text, a text similarity algorithm can be used for semantic comparison. After the comparison is completed, a test verification report is generated that includes the verification results, details of the differences, and the overall pass rate.

[0213] S306. If the comparison results are inconsistent, adjust the target PTC unit in the target workflow and regenerate the target workflow.

[0214] In one possible implementation, the computing device cluster pushes the comparison results to the user, who then manually adjusts the target workflow based on the report content. The user can modify the selection of the target PTC unit, adjust the execution order, correct business parameters, or reconfigure data mapping rules. After the adjustment is completed, S302-S303 are re-executed to generate a new target workflow.

[0215] In another possible implementation, the test execution results include snapshots of the input / output data of each target PTC unit and information on operational anomalies. The computing device cluster uses this information to analyze abnormal nodes in the data flow process and pinpoint the source of the inconsistencies in the comparison.

[0216] Specifically, if the input data format of a unit is found to be inconsistent with expectations, the output interface information of the upstream unit is traced back, and the data mapping rules are automatically corrected; if an execution anomaly is detected in a unit, it is attempted to replace it with another PTC unit with similar functionality; if improper configuration of business parameters is detected, the parameter values ​​are adjusted according to the anomaly information. After the adjustment is completed, the target workflow is automatically regenerated.

[0217] In this way, by testing and executing the target workflow and comparing the results with expectations, an automated feedback loop is formed. Users can use this to verify the effectiveness of the process and adjust deviations in a timely manner, which helps to improve the match between the final workflow and actual business needs and reduces the risk of rework after the process is put into production.

[0218] The above describes the configuration process for the target workflow. The following section details the execution process of the target workflow.

[0219] In an exemplary embodiment, after generating the target workflow, it can be deployed online for execution according to the user's scheduling instructions. This process is as follows: Figure 9 As shown, following S303, the method further includes S307-S308: S307. In response to the user's business scheduling command for the target workflow, based on the identifier of each target PTC unit in the target workflow and the technical parameters of each target PTC unit, create an instantiated object of each target PTC unit.

[0220] In this context, the instantiated object of the target PTC unit refers to the executable entity generated in the runtime environment after allocating computing resources, loading the corresponding business intelligence agent, and injecting technical parameter configurations for each target PTC unit. Each instantiated object has an independent runtime context and can execute business logic and process data according to the unit definition.

[0221] For example, creating instantiated objects for each target PTC unit includes: at the resource level, allocating computing resources, memory space, and runtime threads for each unit; at the code level, loading the business intelligence code or calling interface corresponding to the unit; at the configuration level, injecting the technical parameters determined by S303 into the instantiated objects; and at the link level, establishing the calling relationship and data channel between each instantiated object according to the execution order and data mapping rules defined in the workflow.

[0222] In one possible implementation, after receiving the user's scheduling instruction, the computing device cluster first parses the identifiers of each target PTC unit in the target workflow and loads the corresponding unit definition from the PTC library. Then, it allocates a runtime instance for each unit and injects the technical parameters determined in S303 into the corresponding instantiated object. Simultaneously, based on the execution order and data mapping rules defined in the workflow, it establishes call chains and data channels between the instantiated objects. After completing the above configuration, the instantiated objects of each target PTC unit are in a pending execution state.

[0223] S308. Based on the execution order, the instantiated objects of each target PTC unit are executed sequentially, and the data conversion and transmission between each target PTC unit are processed according to the data mapping rules to complete the business processing process corresponding to the business requirement description information.

[0224] The workflow execution process is a data processing pipeline that proceeds step by step according to a predetermined sequence and rules. After receiving input data, each instantiated object calls its internal business intelligence agent to execute business logic, generate output data, and pass it to downstream units until the entire process is completed.

[0225] In one possible implementation, the computing device cluster sequentially calls the instantiated objects of each target PTC unit according to the execution order determined in S301. When executing each unit, based on the data mapping rules determined in S30141, the required data is obtained from the output of the preceding unit. After field matching, type conversion, and format adjustment according to the mapping rules, it is passed as the input data for the current unit. After the current unit finishes execution, its output data is passed to the subsequent units according to the same mapping rules, until all units have completed execution, and finally, the business processing result is output.

[0226] In some other embodiments, S307-S308 may also be executed after S306 described above.

[0227] In this way, instantiated objects of each PTC unit are created based on the target workflow, and data transformation and transmission are automatically processed according to the execution order. This mechanism clarifies the automated execution path during workflow runtime, freeing users from the need to concern themselves with the underlying scheduling details of the workflow. This helps ensure the accurate execution of business logic and reduces the operational complexity after process deployment.

[0228] In an exemplary embodiment, a management agent can be deployed in the computing device cluster to receive user-input business requirement descriptions, generate initial workflows (such as executing S301 and S3011-S3014), and determine data mapping rules between PTC units in the workflow (such as S30141-S30142), serving as an interactive hub coordinating business intent and technical implementation. This allows business personnel to describe overall requirements in business language without needing to concern themselves with the underlying technical implementation details, further lowering the technical threshold for workflow orchestration.

[0229] For example, the process of managing intelligent agents to perform workflow orchestration includes: Configuration phase: Users input business requirement descriptions through the management agent: "Log in to several email accounts at set times each day to check if they have received product price confirmation emails. If so, log in to the company's internal website to query product-related data, take screenshots and download materials, then use these materials to create emails, and finally send the emails to the corresponding email addresses according to product categories and notify designated personnel in the company's OA system."

[0230] The management agent performs semantic parsing on the business requirement description information, matches four target PTC units from the PTC unit library: "Information Monitoring PTC", "Data Acquisition PTC", "Document Creation PTC", and "Communication Push PTC", and automatically generates an initial workflow topology according to the business logic, establishing the execution order and data mapping relationship between units.

[0231] The management agent coordinates the business agents within each target PTC unit, guiding the user to input execution requirement descriptions for each PTC unit sequentially. For the information monitoring PTC, the user inputs: "The monitoring email addresses are public1@aa.com and public2@aa.com, with passwords of ××× respectively. The emails will be checked every hour from 8:00 AM to 8:00 PM daily, and the subject line must include 'Product Price Confirmation'." The business parameters generated by the PTC's business intelligence agent after parsing include: a list of email accounts, email passwords, monitoring time range, monitoring frequency, and email subject keywords.

[0232] For data acquisition PTC, the user inputs: "The acquisition URL is http: / / internal.aa.com / product. I need to query the product order number, take a screenshot of the page, and download the product specification document." After parsing, the business intelligence agent of data acquisition PTC generates the following business parameters: acquisition URL, source of query parameters (product order number), and acquisition actions (screenshot, download document).

[0233] For document creation PTC, the user inputs: "Use the product price confirmation template in the knowledge base, insert the collected images into the email body, and attach the downloaded document." The business intelligence agent that generates the following business parameters after parsing the data: knowledge base name, template name, image insertion position (body), and document processing method (attachment).

[0234] For push notifications (PTC), the user inputs: "Send Category A products to sales_a@aa.com, Category B products to sales_b@aa.com, and notify the corresponding product managers Zhang San and Li Si in the OA system." The business parameters generated after parsing by the business intelligence agent of the push notification PTC include: the mapping relationship between product category and email address, and the list of personnel to be notified in the OA system.

[0235] Each PTC unit's business intelligence agent parses the execution requirements, generates business parameters, maps them to technical parameters, and completes the internal configuration of each unit.

[0236] After the workflow is built, a trial run is conducted to allow users to verify the execution effect. If errors or unexpected issues arise, the PTC units can be added, deleted, modified, or replaced through the management agent, or parameters can be adjusted through the agents within each PTC until the target workflow that meets the requirements is generated.

[0237] Operation phase: Users will deploy and execute the configured workflow online. The Information Monitoring PTC will periodically monitor designated email addresses according to its configuration. When an email matching the criteria is detected, the product order number will be extracted and transmitted to the Data Acquisition PTC. The Data Acquisition PTC will access the intranet to query, take screenshots, and download the data, then transmit the collected data to the Document Creation PTC. The Document Creation PTC will select a template based on the product order number, generate an email with images and attachments, and send it to the corresponding email address according to the product category. Simultaneously, the Notification and Communication Push PTC will remind designated personnel in the OA system, completing the automated processing of the entire business process.

[0238] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware. The program can be stored in the aforementioned computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be any of the foregoing embodiments or memory. The aforementioned computer-readable storage medium can also be an external storage device of the recovery device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the recovery device. Further, the aforementioned computer-readable storage medium can include both internal storage units of the recovery device and external storage devices. The aforementioned computer-readable storage medium is used to store the aforementioned computer program and other programs and data required by the recovery device. The aforementioned computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0239] This application also provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform any of the methods provided in the above embodiments.

[0240] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0241] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

[0242] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A workflow orchestration method, characterized in that, Applied to computing device clusters; the computing device cluster includes multiple packaged task capability (PTC) units; The PTC unit is a pre-packaged executable task unit; each PTC unit includes: metadata of the PTC unit and a business intelligence agent of the PTC unit, wherein the metadata is used to describe the execution capability characteristics of the corresponding PTC unit; the business intelligence agent is used to execute the business logic of the PTC unit; the method includes: The user's business requirement description information is matched with the metadata of each PTC unit in the plurality of PTC units to generate an initial workflow corresponding to the business requirement description information; the initial workflow includes the identifiers of multiple target PTC units that match the business requirement description information in the plurality of PTC units, as well as the execution order among the multiple target PTC units; In response to the execution requirement description information input by the user for each target PTC unit in the initial workflow, the business intelligence agent of each target PTC unit processes the corresponding execution requirement description information and generates business parameters for each target PTC unit; the business parameters are used to describe the execution information of the corresponding target PTC unit from a business perspective; Based on the service parameters of each target PTC unit and the preset mapping relationship between service parameters and technical parameters, the technical parameters of each target PTC unit are determined, and the target workflow corresponding to the service requirement description information is obtained.

2. The method according to claim 1, characterized in that, The metadata of each PTC unit includes: functional type characteristics, and one or more of execution resource consumption characteristics and execution performance characteristics; the step of matching the user's business requirement description information with the metadata of each PTC unit in the plurality of PTC units to generate an initial workflow corresponding to the business requirement description information includes: Identify the sequence of business actions and the business priority of the business requirement description information; the sequence of business actions includes multiple business actions and the execution order among the multiple business actions. Each of the business actions is semantically matched with the functional type features of each of the PTC units to determine at least one candidate PTC unit corresponding to each of the business actions. In each candidate PTC unit corresponding to the service action, the target PTC unit corresponding to each service action is determined by matching the service priority with the execution resource consumption characteristics and / or execution performance characteristics of each candidate PTC unit. Based on the identifiers of each target PTC unit and the execution order, an initial workflow corresponding to the business requirement description information is generated.

3. The method according to claim 2, characterized in that, The metadata of the PTC unit also includes data interface information, which describes the data structure of the input / output data of the corresponding PTC unit; the generation of the initial workflow corresponding to the business requirement description information based on the identifier of each target PTC unit and the execution order includes: Based on the execution order and the data interface information of each target PTC unit, the data mapping rules between adjacent target PTC units are determined. The initial workflow is generated based on the identifiers of each target PTC unit, the execution order, and the data mapping rules.

4. The method according to any one of claims 1-3, characterized in that, The process of processing the corresponding execution requirement description information through the business intelligence agent of each target PTC unit to generate business parameters for each target PTC unit includes: For any first target PTC unit, the business intelligence agent of the first target PTC unit performs semantic analysis on the execution requirement description information input by the user, and extracts the execution elements of the first target PTC unit from the execution requirement description information; the execution elements include at least the execution target, execution constraints, and data source; Based on each of the execution elements, business parameters are matched in the preset business parameter template to obtain the initial business parameters of the first target PTC unit; each initial business parameter includes: parameter name and recommended parameter value; Display the initial service parameters to the user; In response to the user's editing operation on the initial service parameters, the initial service parameters are updated to obtain the service parameters of the first target PTC unit.

5. The method according to claim 4, characterized in that, The recommended parameter values ​​are obtained in the following way: Extracted from the execution elements; or, Based on the execution element, a preset parameter value matching the execution element is retrieved from a preset domain knowledge base and used as the recommended parameter value; or, Based on the historical value records of the initial service parameters in the first target PTC unit, the historical frequency of each parameter value is counted, and the parameter value with the highest frequency is used as the recommended parameter value.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Perform a test on the target workflow and obtain the test execution results; The expected execution result is determined based on the business requirement description information, and the test execution result is compared with the expected execution result; the expected execution result includes at least one of the expected output format, expected output content, or expected execution indicators; If the comparison results are inconsistent, the target PTC unit in the target workflow is adjusted, and the target workflow is regenerated.

7. The method according to any one of claims 1-6, characterized in that, The computing device cluster also includes multiple historical workflow templates, each of which includes corresponding business metadata. The step of matching the user's business requirement description information with the metadata of each PTC unit to generate the initial workflow corresponding to the business requirement description information includes: Determine the semantic matching degree between the business metadata of each historical workflow template and the business requirement description information; If there is a historical workflow template with a semantic matching degree higher than a preset threshold, the historical workflow template with the highest semantic matching degree will be used as the initial workflow corresponding to the business requirement description information. or, In the absence of a historical workflow template with a semantic matching degree higher than a preset threshold, the user's business requirement description information is matched with the metadata of each PTC unit to generate the initial workflow corresponding to the business requirement description information.

8. The method according to claim 1, characterized in that, The target workflow includes data mapping rules between each target PTC unit; the method further includes: In response to the user's business scheduling command for the target workflow, based on the identifier of each target PTC unit in the target workflow and the technical parameters of each target PTC unit, an instantiated object of each target PTC unit is created; Based on the execution order, the instantiation objects of each target PTC unit are executed sequentially, and the data conversion and transmission between each target PTC unit are processed according to the data mapping rules to complete the business processing process corresponding to the business requirement description information.

9. The method according to any one of claims 1-8, characterized in that, The PTC unit includes at least: a basic PTC unit and a domain PTC unit; the basic PTC unit is an atomic PTC unit; the domain PTC unit is composed of the basic PTC unit and / or other domain PTC units; the domain PTC unit is configured with a knowledge base, which is used to provide the business agents in the domain PTC unit with the domain knowledge required for business execution.

10. A computing device cluster, characterized in that, The computing device cluster includes at least one computing device, which includes memory and a processor. The memory is used to store program instructions; The processor is used to execute the program instructions, causing the computing device cluster to perform the workflow orchestration method as described in any one of claims 1-9.

Images