Task processing method and device based on agent canvas arrangement, equipment and medium

Abstract

The invention belongs to the field of artificial intelligence, and relates to a task processing method and device based on agent arrangement canvas, equipment and a medium, and the method comprises the steps: receiving task information of a target task input by a user on the agent arrangement canvas, and splitting a task target in the task information into sub-task chains by using a semantic analysis model built in the canvas; obtaining a corresponding target agent from the agent capability library, generating a recommendation list, and constructing a preliminary arrangement process; a conflict judgment model is used for detecting resource conflicts, if the resource conflicts exist, a conflict detector generates a resolution strategy based on execution constraint and task core degree, and the conflicts are processed to obtain a conflict-free process; and optimizing the execution sequence of the conflict-free process according to the operation performance index and the task core degree of each agent to obtain a target arrangement process, and executing the process to output a target task execution result. The method can be applied to the business fields of financial science and technology, insurance, medical treatment and the like, and the efficiency of multi-agent cooperative execution of complex tasks and the task execution reliability can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and is applied to online processing business scenarios such as fintech, insurance, and healthcare. In particular, it relates to a task processing method, device, equipment, and medium based on intelligent agent orchestration canvas. Background Technology

[0002] With the large-scale deployment of large-scale modeling technology in enterprise-level scenarios, multi-agent collaboration has become a core mode for handling complex business tasks. Specialized agents such as data extraction, report generation, and risk assessment collaborate to achieve efficient progress in business processes. In financial business scenarios, multi-agent collaboration is widely used in key business processes such as auto insurance claims settlement, medical insurance underwriting, credit risk assessment, and intelligent investment advisory strategy generation. Enterprises have an increasingly urgent need for multi-agent collaboration, but current related technical solutions still have many shortcomings and are unable to meet the intelligent and precise requirements of practical applications.

[0003] In existing technologies, multi-agent collaboration lacks deep semantic parsing capabilities for tasks, failing to automatically decompose high-level user-proposed goals into a logically coherent sequence of sub-tasks. This necessitates manual decomposition and configuration, which is not only cumbersome but also prone to impacting overall execution performance due to flawed decomposition logic. Furthermore, the system struggles to accurately identify the execution characteristics and relationships of different agents, failing to automatically match agents to suitable sub-tasks or predict input-output dependencies between agents, resulting in a lack of scientific execution sequence planning in the collaborative process. Moreover, existing solutions lack effective automated detection and resolution mechanisms for resource conflicts such as accessing the same data source or writing to the same target field, often requiring manual intervention. This is not only inefficient but can also lead to data corruption and task interruptions. These technical deficiencies in financial business scenarios directly result in extended auto insurance claims processing times, decreased underwriting accuracy, and increased credit risk assessment bias, leading to low efficiency and poor reliability when multi-agent collaboratively executes complex tasks. Summary of the Invention

[0004] The purpose of this application is to propose a task processing method, apparatus, computer device and storage medium based on an intelligent agent orchestration canvas, so as to solve the problems of low efficiency and poor reliability of task execution when multiple intelligent agents cooperate to perform complex tasks.

[0005] Firstly, a task processing method based on intelligent agent-based canvas arrangement is provided, which adopts the following technical solution: The system receives task information of the target task input by the user in the agent orchestration canvas; it uses the semantic analysis model built into the agent orchestration canvas to break down the task objective in the task information into sub-task chains; it retrieves the target agents corresponding to the sub-task chains from the preset agent capability library, generates an agent recommendation list, and constructs a preliminary orchestration process based on the agent recommendation list; it uses a preset conflict judgment model to detect whether there are resource conflicts between the target agents in the preliminary orchestration process; if resource conflicts exist, it uses a preset conflict detector to generate a conflict resolution strategy based on the execution constraints in the task information and the task core degree of each target agent; based on the conflict resolution strategy, it performs conflict resolution processing on the preliminary orchestration process to obtain an orchestration process without resource conflicts; based on the running performance indicators and task core degree of each target agent, it optimizes the execution order of the orchestration process without resource conflicts to obtain the target orchestration process; it executes the target orchestration process to output the execution result of the target task.

[0006] Secondly, a task processing device based on an intelligent agent orchestration canvas is provided, which adopts the following technical solution: The receiving module is used to receive task information of the target task input by the user in the intelligent agent arrangement canvas; The splitting module is used to split the task objectives in the task information into sub-task chains by using the semantic analysis model built into the intelligent agent orchestration canvas. The acquisition module is used to acquire the target intelligent agents corresponding to the sub-task chain from the preset intelligent agent capability library, generate an intelligent agent recommendation list, and construct a preliminary orchestration process based on the intelligent agent recommendation list; The detection module is used to detect whether there are resource conflicts between the target agents in the initial orchestration process using a preset conflict judgment model. The generation module is used to generate conflict resolution strategies based on the execution constraints in the task information and the task core degree of each target agent if resource conflicts exist, by using a preset conflict detector. The processing module is used to resolve conflicts in the initial orchestration process based on the conflict resolution strategy, so as to obtain an orchestration process without resource conflicts. The optimization module is used to optimize the execution order of orchestration processes without resource conflicts based on the operational performance indicators and task coreness of each target agent, to obtain the target orchestration process, execute the target orchestration process, and output the execution result of the target task.

[0007] Thirdly, a computer device is provided, which adopts the following technical solution: The system receives task information of the target task input by the user in the agent orchestration canvas; it uses the semantic analysis model built into the agent orchestration canvas to break down the task objective in the task information into sub-task chains; it retrieves the target agents corresponding to the sub-task chains from the preset agent capability library, generates an agent recommendation list, and constructs a preliminary orchestration process based on the agent recommendation list; it uses a preset conflict judgment model to detect whether there are resource conflicts between the target agents in the preliminary orchestration process; if resource conflicts exist, it uses a preset conflict detector to generate a conflict resolution strategy based on the execution constraints in the task information and the task core degree of each target agent; based on the conflict resolution strategy, it performs conflict resolution processing on the preliminary orchestration process to obtain an orchestration process without resource conflicts; based on the running performance indicators and task core degree of each target agent, it optimizes the execution order of the orchestration process without resource conflicts to obtain the target orchestration process; it executes the target orchestration process to output the execution result of the target task.

[0008] Fourthly, a computer-readable storage medium is provided, which adopts the following technical solution: The system receives task information of the target task input by the user in the agent orchestration canvas; it uses the semantic analysis model built into the agent orchestration canvas to break down the task objective in the task information into sub-task chains; it retrieves the target agents corresponding to the sub-task chains from the preset agent capability library, generates an agent recommendation list, and constructs a preliminary orchestration process based on the agent recommendation list; it uses a preset conflict judgment model to detect whether there are resource conflicts between the target agents in the preliminary orchestration process; if resource conflicts exist, it uses a preset conflict detector to generate a conflict resolution strategy based on the execution constraints in the task information and the task core degree of each target agent; based on the conflict resolution strategy, it performs conflict resolution processing on the preliminary orchestration process to obtain an orchestration process without resource conflicts; based on the running performance indicators and task core degree of each target agent, it optimizes the execution order of the orchestration process without resource conflicts to obtain the target orchestration process; it executes the target orchestration process to output the execution result of the target task.

[0009] Compared with existing technologies, the embodiments of this application have the following main advantages: Through multi-stage collaborative design, the core defects of existing multi-agent collaborative technologies are specifically addressed, significantly improving the efficiency and reliability of complex task execution. Utilizing the semantic analysis model built into the agent orchestration canvas, target tasks are automatically decomposed, transforming high-level tasks into logically coherent sub-task chains. This eliminates the need for manual decomposition, simplifying the operation process and ensuring the rationality of the decomposition, laying the foundation for collaborative execution. Through the agent capability library, target agents are automatically matched and adapted based on the sub-task chains to construct a preliminary orchestration process, accurately identifying agent execution characteristics and relationships, avoiding the subjectivity and inefficiency of manual matching. By leveraging the collaboration of a conflict judgment model and a conflict detector, automated detection and resolution of resource conflicts are achieved. Customized strategies are generated by combining task execution constraints and agent task coreity, effectively avoiding problems such as data corruption and task interruption. Simultaneously, the execution order is optimized based on agent performance indicators and task coreity, ensuring the scientific and efficient nature of the process, comprehensively improving the intelligence level and execution reliability of multi-agent collaborative processing of complex tasks. Attached Figure Description

[0010] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 A flowchart of an embodiment of the task processing method based on an intelligent agent orchestration canvas according to this application; Figure 3 This is a schematic diagram of a structure of an embodiment of a task processing device based on an intelligent agent orchestration canvas according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0013] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0014] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0015] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0016] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0017] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers.

[0018] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0019] It should be noted that the task processing method based on intelligent agent orchestration canvas provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the task processing device based on intelligent agent orchestration canvas is generally set in the server / terminal device.

[0020] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0021] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of a task processing method based on an agent-based canvas orchestration according to this application. This agent-based canvas orchestration task processing method includes the following steps: Step S201: Receive the task information of the target task input by the user in the intelligent agent arrangement canvas.

[0022] The Agent Orchestration Canvas is an integrated operating platform that combines task input interaction, semantic parsing and processing, agent matching and filtering, visual orchestration of collaborative processes, and automated conflict handling. It provides end-to-end support for multi-agent collaborative execution of complex business tasks. It can incorporate core modules such as semantic analysis models and agent capability libraries, allowing users to visually configure task logic and adjust the execution order of agents.

[0023] Among them, the target task refers to the high-level business request put forward by the user based on the specific business scenario requirements, which needs to be completed by multiple intelligent agents through division of labor and cooperation. It is the core orientation and ultimate goal of the entire collaborative orchestration process. For example, "generate a market share analysis report for a certain product over the past six months" or "complete a full-dimensional credit risk assessment of new customers and output rating results".

[0024] The task information is a complete set of key data related to the target task that the user inputs in the intelligent agent orchestration canvas. Specifically, it can include core content such as task objectives, specified data sources, output requirements (such as file format and data precision) and execution constraints (such as time limits and permission scope).

[0025] Step S202: Using the semantic analysis model built into the intelligent agent orchestration canvas, the task objective in the task information is split into a chain of sub-tasks.

[0026] The semantic analysis model is an artificial intelligence algorithm model embedded in the intelligent agent orchestration canvas. It possesses deep semantic understanding and business logic decomposition capabilities. By parsing the natural language expressions in task information, it uncovers the core demands and business-related logic of the task objectives, enabling the automated decomposition of high-level task objectives. For example, it can break down "analyzing the market competitiveness of a new product" into a model of coherent sub-tasks such as "competitive product data collection - product parameter comparison - market feedback collection - competitiveness scoring - report generation".

[0027] Among them, the task objective is the core execution guide and final result requirement of the target task. It clarifies the specific effects, output standards or business value that the task needs to achieve. It is the core object of the semantic analysis model for task decomposition and the final destination of the entire collaborative process.

[0028] The subtask chain is a sequence of subtasks with clear logical connections and execution order, formed by the semantic analysis model based on the decomposition of the task objective. Each subtask corresponds to a specific business link in the objective task, and the output of the previous subtask provides support for the input of the next subtask, forming a closed-loop collaborative logic.

[0029] Step S203: Obtain the target intelligent agent corresponding to the sub-task chain from the preset intelligent agent capability library, generate an intelligent agent recommendation list, and construct a preliminary orchestration process based on the intelligent agent recommendation list.

[0030] The intelligent agent capability library is a pre-built structured database that stores the core attribute information of various candidate intelligent agents. It can include key information such as the capability description of each candidate intelligent agent (e.g., data extraction, risk assessment, report rendering), data interaction specifications (e.g., input / output data formats, interface protocols), and access resource scope (e.g., accessible data sources, callable tools). It is used to provide comprehensive and accurate data support for matching and adapting intelligent agents to each subtask in the subtask chain.

[0031] The target agent is a candidate agent selected from the agent capability library through a matching algorithm. These agents possess core capabilities that highly match the execution requirements of the sub-tasks, enabling them to undertake and efficiently complete the corresponding sub-tasks. The agent recommendation list is a set of target agents generated based on the matching degree analysis results between sub-task requirements and agent capabilities, output from the agent capability library after multi-dimensional matching calculations. The initial orchestration process is an initial multi-agent collaborative execution framework constructed based on the sub-task chain logic and the agent recommendation list, after calculating agent collaboration relationships.

[0032] Step S204: Using a preset conflict judgment model, detect whether there are resource conflicts between the target intelligent agents in the preliminary arrangement process.

[0033] The conflict detection model is an algorithm used to detect resource access conflicts among agents in the initial orchestration process. Its core is to extract the resource access sets of each target agent, including read and write resource sets, and perform matching analysis on the agents' resource access behaviors to output a conflict detection result dataset. Based on this dataset, it can determine whether write conflicts, read conflicts, or read-write conflicts exist. For example, if two target agents are detected simultaneously writing to the same database field or concurrently reading from the same restricted data source, a conflict detection result is output.

[0034] Resource conflict refers to a conflicting state in the initial orchestration process where resource access behaviors of any two or more target agents contradict each other. Specifically, this can include write conflicts and read conflicts. A write conflict occurs when multiple target agents simultaneously perform write operations on the same resource, such as a database field or file path; a read conflict occurs when multiple target agents concurrently access the same restricted resource, such as a data source or API interface that does not support concurrency. Such conflicts can lead to data corruption, access blockages, and other problems, requiring conflict resolution strategies.

[0035] In step S205, if resource conflicts exist, a conflict resolution strategy is generated based on the execution constraints in the task information and the task core degree of each target agent through a preset conflict detector.

[0036] The conflict detector is a core processing module that integrates conflict data extraction, coreity calculation, and policy matching functions to generate conflict resolution strategies. For example, it's a module that generates time-sharing scheduling strategies for data source access conflicts. Execution constraints are explicit rules and conditions in the task information that restrict the execution of collaborative processes. They originate from user business needs or system limitations and can include time limits, permission scope, data format requirements, etc.

[0037] Task coreness is a quantitative indicator that represents the importance of the subtasks corresponding to the target agent within the target task. It is calculated using a preset algorithm that combines the relationship between the subtask chain and the target task. Its value determines the resource allocation and execution priority during conflict resolution. For example, in "financial statement generation," the task coreness of the data accounting agent is higher than that of the formatting agent.

[0038] Among them, the conflict resolution strategy is a specialized solution generated by the conflict detector for resource conflicts identified in the initial orchestration process. Based on the original execution sequence of the initial orchestration process, it first extracts the conflict type from the conflict detection results, combines it with the execution constraints in the task information, and then incorporates the task core degree of each target agent calculated by a preset algorithm. It then calls a preset conflict handling mapping library to match and determine specific strategies according to different conflict scenarios (such as concurrent access to data sources, competition for writing target fields, etc.), which may include priority scheduling, time-sharing resource access, and agent execution sequence adjustment.

[0039] Step S206: Based on the conflict resolution strategy, perform conflict resolution processing on the preliminary orchestration process to obtain an orchestration process without resource conflicts.

[0040] Among them, conflict resolution is a specific operational process that makes targeted adjustments to the initial orchestration process based on the conflict resolution strategy. By modifying the execution order of agents, resource access methods, data interaction paths, or replacing conflicting agents, resource conflicts in the initial orchestration process can be completely eliminated.

[0041] Among them, the resource-free orchestration process is a collaborative execution process that completely eliminates resource competition and execution logic conflicts between intelligent agents after conflict resolution. Each target intelligent agent can smoothly collaborate according to the established logic in this process without the risk of data corruption or execution interruption.

[0042] Step S207: Based on the operational performance indicators and task coreness of each target agent, optimize the execution order of the orchestration process without resource conflicts to obtain the target orchestration process, execute the target orchestration process, and output the execution result of the target task.

[0043] Among them, performance metrics are a set of quantitative parameters characterizing the agent's execution capabilities, efficiency, and stability, obtained by statistically analyzing historical execution data. These metrics may include usage frequency (number of historical calls), task complexity suitability (success rate in handling complex tasks), execution success rate (success rate in completing subtasks), average execution time (average time to complete a task), and resource consumption rate (resource usage ratio during execution). These metrics are crucial for optimizing the execution order of orchestration processes without resource conflicts.

[0044] Among them, the target orchestration process is a final collaborative execution scheme formed by adjusting the execution order through a multi-objective optimization algorithm, based on the resource-free orchestration process and combining the operational performance indicators of each target agent with the task coreness of the sub-tasks.

[0045] The execution results are the complete data set output after the target orchestration process is fully executed according to the established logic. This includes not only the core business results (i.e., task results) that meet the task information output requirements, but also key process data from the collaborative processes of various agents. Its core purpose is to provide a data source for subsequent parsing and extraction of task results and execution logs. The execution logs can be further used to mine information such as agent dependencies and conflict handling records, providing training data for the iterative optimization of semantic analysis models, agent capability libraries, and conflict detectors.

[0046] This application's embodiments address the core shortcomings of existing multi-agent collaborative technologies through multi-stage collaborative design, significantly improving the efficiency and reliability of complex task execution. Leveraging the semantic analysis model built into the agent orchestration canvas, target tasks are automatically decomposed, transforming high-level tasks into logically coherent sub-task chains. This eliminates the need for manual decomposition, simplifying the process and ensuring the rationality of the decomposition, laying the foundation for collaborative execution. Through an agent capability library, target agents are automatically matched and adapted based on the sub-task chains to construct a preliminary orchestration process, accurately identifying agent execution characteristics and relationships, avoiding the subjectivity and inefficiency of manual matching. The collaboration of a conflict judgment model and a conflict detector enables automated detection and resolution of resource conflicts. Combined with task execution constraints and agent task coreity, customized strategies are generated, effectively avoiding data corruption and task interruptions. Simultaneously, the execution order is optimized based on agent performance indicators and task coreity, ensuring the scientific and efficient nature of the process and comprehensively improving the intelligence level and execution reliability of multi-agent collaborative processing of complex tasks.

[0047] In some optional implementations of this embodiment, step 203, obtaining the target intelligent agent corresponding to the sub-task chain from a preset intelligent agent capability library and generating an intelligent agent recommendation list, specifically includes the following steps: The system invokes the built-in agent capability library in the agent orchestration canvas. This library stores the capability descriptions, data interaction specifications, and access resource ranges for each candidate agent. Based on these descriptions, specifications, and resource ranges, a capability feature vector is generated for each candidate agent. A semantic analysis model is used to generate a task feature vector for each subtask in the subtask chain. The matching degree between the capability feature vector and each task feature vector is calculated. Candidate agents with matching degrees greater than a preset threshold are selected as target agents, generating a recommended agent list.

[0048] Among them, the capability feature vector is a multi-dimensional numerical vector generated by quantizing the core attribute information of candidate intelligent agents in the intelligent agent capability library. It is used to accurately characterize the execution capability and adaptation characteristics of the intelligent agent.

[0049] The task feature vector is a multi-dimensional numerical vector generated by the semantic analysis model after parsing each subtask in the task chain. It represents the core requirements of the subtasks. It is constructed based on key information such as the business objectives, data processing requirements, and resource access requirements of the subtasks, transforming the unstructured requirements of the subtasks into a standardized numerical form.

[0050] In one example, within a customer credit risk assessment scenario, the user inputs the target task "to conduct credit risk rating for applicant companies" through the agent orchestration canvas. This task is then broken down into a sub-task chain: "enterprise financial data extraction - credit information verification - risk indicator calculation - rating result generation" using a semantic analysis model. The agent orchestration canvas's built-in agent capability library is invoked. This library stores capability descriptions (e.g., "supports extraction of financial statement data from listed companies"), data interaction specifications (e.g., "outputs structured data in JSON format"), and access resource scope (e.g., "can access the bank's internal credit database") for candidate agents such as financial data extraction agents and credit verification agents. Based on this information, the financial data extraction agent is transformed into a capability feature vector containing dimensions such as "financial data processing, JSON output, and access to the bank's credit database." The semantic analysis model then generates a task feature vector for the "enterprise financial data extraction" sub-task, containing "financial data collection, structured output, and access to the credit database." The matching degree between the two is calculated. If the result is 0.92 (higher than the preset threshold of 0.8), the financial data extraction agent is included in the recommendation list. Similarly, complete the matching of other sub-tasks with intelligent agents, and generate an intelligent agent recommendation list that includes intelligent agents for credit verification and intelligent agents for risk indicator calculation, ensuring that each sub-task is matched with a target intelligent agent with suitable capabilities.

[0051] This application's embodiments precisely address the problems of poor adaptability and low matching efficiency between agents and subtasks in existing technologies through a collaborative design of agent capability library and feature vector matching. Leveraging the complete attribute information stored in the agent capability library, comprehensive data support is provided for matching. Furthermore, unstructured information such as capability descriptions and data interaction specifications are transformed into standardized capability feature vectors. Simultaneously, a semantic analysis model is used to generate task feature vectors for subtasks, achieving quantitative matching between agents and subtasks. By calculating the matching degree and selecting agents above a certain threshold to generate a recommendation list, the subjectivity and blindness of manual matching are avoided, ensuring a high degree of fit between the target agent's capabilities and the subtask requirements.

[0052] In some optional implementations, step 203, constructing a preliminary orchestration process based on the agent's recommendation list, specifically includes the following steps: Using a semantic analysis model, the task information is integrated with the capability descriptions, data interaction specifications, and access resource scope of each target agent in the agent recommendation list to generate capability embedding vectors for each target agent. Based on the capability embedding vectors, the input-output similarity between adjacent target agents and the resource-task similarity between each target agent and the target task are calculated. If both the input-output similarity and the resource-task similarity are greater than a preset threshold, dependency connections between each target agent are established in the agent orchestration canvas to generate a preliminary orchestration process.

[0053] Integration refers to the process by which the semantic analysis model multi-dimensionally correlates and fuses the task information (including task objectives, specified data sources, output requirements, and execution constraints) with the capability descriptions, data interaction specifications, and access resource scope of the target intelligent agents in the recommended agent list. This process uncovers the business logic and compatibility relationships between the information, providing complete and relevant foundational data for subsequent vector generation. The capability embedding vector is a multi-dimensional numerical vector generated after feature extraction and quantification of the integrated information. It encompasses core attributes such as the intelligent agent's capability boundaries, data interaction standards, and resource access permissions, accurately representing the compatibility characteristics and collaborative potential between the target intelligent agent and the task.

[0054] Input-output similarity refers to the degree of fit between the output data format and content dimensions of the preceding target agent and the input requirements of the subsequent target agent. This indicator directly reflects the compatibility of data flow between agents and is a key basis for judging whether agents can connect smoothly. Resource-task similarity refers to the degree of matching between the target agent's access to resources and resource access permissions and the data source requirements and resource usage requirements of the target task. This indicator ensures that the agent has the resource conditions to complete the corresponding sub-task and is a core reference for selecting suitable agents.

[0055] The preset threshold is a similarity benchmark value pre-set based on historical collaboration data and business scenario requirements. Collaboration between agents is only deemed feasible when both input-output similarity and resource-task similarity exceed this value. Dependency connections are visual logical association markers established in the agent orchestration canvas. These markers characterize the execution order and data flow path between target agents and are the core carrier for visualizing the initial orchestration process.

[0056] In one example, in a car insurance claims risk verification scenario, the objective task is "to complete the fraud risk assessment of car insurance accident claims," ​​and the sub-task chain is "accident data collection - damage assessment report verification - historical claims query - risk level assessment." The recommended agent list includes target agents such as accident data collection agents and damage assessment verification agents. A semantic analysis model is invoked to integrate task information (specifying the car insurance database as the data source, outputting a risk level report, and completion within 24 hours) with the capability descriptions of each target agent, JSON-formatted interaction specifications, and car insurance database access permissions, generating capability embedding vectors for each agent. Based on these vectors, the input-output similarity of adjacent agents is calculated. For example, the similarity between the output of the data collection agent and the input of the damage assessment verification agent is 0.9. Simultaneously, the resource task similarity between each agent and the task is calculated to be 0.88, both exceeding the preset threshold of 0.8. Dependency connections between agents are established in the agent orchestration canvas, generating a preliminary orchestration process of "accident data collection agent - damage assessment verification agent - historical claims query agent - risk level assessment agent."

[0057] This application's embodiments address the problem of unscientific agent collaborative logic planning in existing technologies through a hierarchical design involving information integration, vector generation, and similarity determination. By integrating task information with core agent attributes using a semantic analysis model, it generates capability embedding vectors covering ability, interaction, and resource dimensions, achieving a quantitative representation of task and agent characteristics. By calculating the input-output similarity of neighboring agents and the resource-task similarity between agents and tasks, and combining this with preset thresholds to filter adaptation relationships and establish dependency connections, the subjectivity of manual planning is avoided. This design ensures that the agent connections in the initial orchestration process are supported by data, laying a scientific foundation for subsequent conflict detection and process optimization, and improving the rationality and automation level of multi-agent collaborative processes.

[0058] In some optional implementations, the step "calculating the input-output similarity between adjacent target agents and the resource-task similarity between each target agent and the target task based on the capability embedding vector" specifically includes the following steps: For each capability embedding vector, it is split into input feature sub-vectors, output feature sub-vectors, and resource requirement sub-vectors. For all target agents, the similarity between the output feature sub-vectors of adjacent preceding target agents and the input feature sub-vectors of adjacent subsequent target agents is calculated to obtain the input-output similarity between adjacent target agents. The task information is parsed to obtain the resource requirement feature vector of the target task. The similarity between the resource requirement sub-vector and the resource requirement feature vector of each target agent is calculated to obtain the resource task similarity between each target agent and the target task.

[0059] The input feature sub-vector is a multi-dimensional numerical vector extracted from the capability embedding vector of the target agent, representing its input data requirements. It encompasses key attributes such as input data format, data dimension, and data precision, and is used to determine whether the output of the preceding agent matches its input requirements. The output feature sub-vector is a multi-dimensional numerical vector extracted from the capability embedding vector, representing the characteristics of the agent's output data. It includes information such as output data type, data fields, and output frequency, providing a quantitative basis for matching the input of subsequent agents. The resource requirement sub-vector is a multi-dimensional numerical vector extracted from the capability embedding vector, representing the resources required by the agent to perform its task. It involves data source access permissions, hardware resources, and interface call permissions, reflecting the agent's resource dependencies.

[0060] Among them, the resource demand feature vector is a multi-dimensional numerical vector generated by parsing task information, which represents the overall resource demand of the target task. It integrates information such as the task-specified data source, resource usage restrictions, and key resource priorities, and is used to verify the matching degree between the agent's resource demand and the task demand.

[0061] In one example, within an intelligent underwriting scenario for insurance policies, the objective is to "automatically underwrite life insurance applications." The target agents include an application information collection agent, a health declaration verification agent, and a risk level calculation agent. The capability embedding vectors of each agent are decomposed into input, output, and resource requirement feature sub-vectors. For instance, the input sub-vector of the health declaration verification agent represents "receiving structured application information," the output sub-vector represents "outputting verification results and anomaly annotations," and the resource sub-vector represents "accessing the health declaration database." The input-output similarity between adjacent agents is calculated; the similarity between the output sub-vector of the application information collection agent and the input sub-vector of the health declaration verification agent is 0.91. Parsing the task information yields a resource requirement feature vector, which includes "accessing the application information database and the health declaration database." The similarity between the resource sub-vector of the health declaration verification agent and this vector is 0.89. These two similarity results provide the core basis for subsequent determination of agent dependency connections and the generation of preliminary orchestration processes.

[0062] This application's embodiments achieve refined quantitative representation of agent characteristics by decomposing the capability embedding vector into three sub-vectors: input, output, and resource requirements. For adjacent agents, the similarity between the preceding output and subsequent input feature sub-vectors is calculated to accurately determine the data flow compatibility between agents. By parsing task information to generate resource requirement feature vectors, and then calculating their similarity with the agent's resource requirement sub-vectors, the similarity between the agent's resource conditions and task requirements is ensured. This design, through hierarchical similarity calculation, provides accurate data support for subsequent dependency establishment, avoids the subjectivity of manual judgment, and significantly improves the scientific rigor and adaptability of the initial orchestration process.

[0063] In some optional implementations, step S204, using a preset conflict judgment model, detects whether there are resource conflicts between the target agents in the initial orchestration process, specifically including the following steps: The access resource set of each target agent is extracted from the initial orchestration process. The access resource set includes a read resource set and a write resource set. Based on the read resource set and the write resource set, the resource access behavior of each target agent in the initial orchestration process is matched and detected using a preset conflict judgment model, and a conflict detection result dataset is output. Based on the conflict detection result dataset, if there is a write conflict, a read conflict, or a read-write conflict between any two target agents, it is determined that there is a resource conflict between the target agents in the initial orchestration process. If there is no write conflict, read conflict, or read-write conflict among all target agents, it is determined that there is no resource conflict among the target agents in the initial orchestration process.

[0064] The access resource set is extracted from the initial orchestration process and comprises all resources that the target agent needs to operate on during task execution. It covers all resources for both reading and writing data, providing a complete set of resource analysis objects for the conflict resolution model. The read resource set is a subset of the access resource set, containing resources that the target agent needs to read during task execution, such as database tables, file storage paths, and API return data. The write resource set is another subset of the access resource set, containing resources that the target agent needs to write during task execution, such as result storage tables, log files, and output documents. For example, a risk rating agent might need to write customer risk profile tables containing risk level results.

[0065] Resource access behavior refers to the actions of a target agent in reading and writing to a resource set, including access time window, operation type (read / write), and resource occupancy duration. It is the core analysis object of the conflict detection model. The conflict detection result dataset is a structured data set output by the conflict detection model after matching and detecting resource access behavior. It contains information such as whether a conflict exists, the identifier of the conflicting agent, the type of conflicting resource, and the conflict time window, providing a basis for subsequent determination of whether a resource conflict exists.

[0066] In one example, in a medical expense insurance claims settlement scenario, the initial orchestration process includes a medical expense collection agent, a medical insurance policy verification agent, and a claims amount calculation agent. The resource sets accessed by each agent are extracted from the process: the medical expense collection agent reads the hospital HIS system expense data table and writes to the original claims data table; the medical insurance policy verification agent reads the medical insurance policy database and the original claims data table, but has no writing resource set. A conflict detection model is used to detect resource access behavior, revealing that both the medical expense collection agent and the medical insurance policy verification agent need to read the original claims data table, and their access time windows overlap. The output conflict detection result dataset is labeled "There is a read conflict between the two agents," indicating that the initial orchestration process has resource conflicts that need further resolution.

[0067] This application addresses the difficulty in identifying resource conflicts in existing technologies by splitting resource sets, using model detection, and defining clear judgment rules. First, it extracts access resource sets containing subsets of read and write resources from the initial orchestration process, fully covering the scope of agent resource operations. Then, based on these two types of resource sets, a conflict judgment model matches and detects resource access behaviors, outputting a structured conflict detection result dataset. Finally, based on the dataset, it clearly determines whether a write or read conflict exists, avoiding the inefficiency and omissions of manual investigation, providing accurate evidence for subsequent conflict resolution, and ensuring the reliability of multi-agent collaboration.

[0068] In some optional implementations, step S205 involves generating a conflict resolution strategy based on the execution constraints in the task information and the task coreness of each target agent using a preset conflict detector. This strategy specifically includes the following steps: Extract the original execution sequence and conflict type of the preliminary orchestration process from the conflict detection result dataset; calculate the task core degree of each target agent based on the association between the sub-task chain and the target task using a preset algorithm; call the preset conflict resolution mapping library, and determine the conflict resolution strategy based on the original execution sequence, combined with the conflict type, execution constraints and task core degree.

[0069] The original execution sequence refers to the initial execution order and time plan of each target agent in the preliminary orchestration process extracted from the conflict detection result dataset. It includes the planned start time, execution duration, data flow nodes, and dependencies of each agent. The conflict type is the specific category of resource conflict identified in the conflict detection results. Based on the resource operation behavior of the agents, it mainly includes three categories: read conflict, write conflict, and read-write conflict. Different types correspond to different resolution logics.

[0070] Among them, the preset algorithm refers to the preset mathematical model or logical algorithm used to calculate the task core degree of the target intelligent agent, such as the weighted summation algorithm or the analytic hierarchy process. The association relationship refers to the business logic relationship between each subtask in the subtask chain and the target task, covering the direct impact of the subtask on the achievement of the target task, and the support and dependency relationships between subtasks. The conflict resolution mapping library is a pre-built structured database that stores the mapping relationship between multi-dimensional condition combinations such as conflict type, execution constraints, and task core degree, and corresponding resolution strategies. For example, the combination of "read conflict + high core degree priority execution + 8-hour time limit constraint" maps to the strategy of "high core degree intelligent agents access first, and other intelligent agents queue in a time-sharing manner."

[0071] In one example, in a critical illness insurance claim review scenario, the initial orchestration process encounters a read conflict between the "medical record data extraction agent" and the "expense compliance verification agent" (both access the hospital's electronic medical record database). The original execution sequence (medical record extraction agent starts at 9:00 AM, expense verification agent starts at 9:05 AM) and conflict type (read conflict) are extracted from the conflict detection result dataset. Using a weighted summation algorithm, based on the association between the sub-task chain (medical record extraction - expense verification - claim review) and the target task, the coreity of the medical record extraction agent is calculated to be 0.8, and the coreity of the expense verification agent is 0.6. The conflict resolution mapping library is invoked, and based on the original execution sequence, combined with the read conflict type, the "review completed within 2 hours" execution constraint, and the coreity, a resolution strategy is determined: "Medical record extraction agent prioritizes access (9:00-9:30 AM), expense verification agent starts at 9:30 AM."

[0072] This application addresses the shortcomings of existing technologies, such as reliance on manual intervention and lack of precise data in conflict resolution, by extracting core conflict information, quantifying task coreness, and matching mapping strategies. First, the original execution sequence and conflict type are extracted from the conflict detection result dataset to clarify the conflict adjustment benchmark and category. Then, a pre-defined algorithm, combined with the relationship between sub-task chains and target tasks, transforms business logic into quantifiable task coreness, avoiding subjective priority determination. Finally, based on the original execution sequence, a conflict resolution mapping library is invoked, integrating conflict type, execution constraints, and task coreness to determine the strategy, ensuring the resolution solution adapts to the scenario requirements and improving the automation and accuracy of conflict handling.

[0073] In some optional implementations, step 207, based on the operational performance indicators and task coreness of each target agent, optimizes the execution order of the orchestration process without resource conflicts to obtain the target orchestration process, specifically including the following steps: The system obtains the operational performance metrics of each target agent from the agent capability library; based on the task coreness and operational performance metrics, it calculates the priority weight of each target agent according to a preset weighting algorithm; based on the priority weight and the dependency relationship between each target agent, it optimizes the execution order of the orchestration process without resource conflicts and generates the target orchestration process.

[0074] Among them, the operational performance indicators are a set of quantitative data reflecting the execution efficiency and stability of the target intelligent agent, obtained from the intelligent agent capability library. These indicators can cover core dimensions such as average response time, task success rate, resource utilization, and fault recovery time, providing objective performance-level basis for priority weight calculation. Conflict type refers to the specific category of resource conflicts determined in the early conflict detection phase, which can include read conflicts, write conflicts, and read-write conflicts. During the execution order optimization phase, these must be used as constraints to ensure that the adjusted process does not trigger the same type of resource conflict again.

[0075] The preset weighting algorithm is a pre-defined mathematical model that integrates task coreness and operational performance indicators to calculate priority. Weight coefficients can be flexibly configured according to different business scenarios. Priority weights are quantitative values ​​representing the execution priority of a target intelligent agent within the process, calculated using the preset weighting algorithm. Higher values ​​indicate that the agent requires priority scheduling. Dependencies are pre- and post-requirement relationships formed between target intelligent agents based on data flow and business logic. For example, "Medical record data extraction must be completed before cost compliance verification can be carried out." Optimizing the execution order must prioritize satisfying this relationship; the inherent logic of the process should not be broken in pursuit of priority.

[0076] In one example, in the auto insurance claims settlement scenario within the financial insurance sector, the initial orchestration process without resource conflicts is "Accident Image Recognition Agent - Repair Quotation Agent - Claims Calculation Agent". Performance metrics are obtained from the agent capability library, including: image recognition agent response time 8 seconds / case, success rate 98.5%; repair quotation agent response time 12 seconds / case, success rate 99.2%; claims calculation agent response time 5 seconds / case, success rate 99.8%. Known task coreity: image recognition agent task coreity is 0.9, repair quotation agent task coreity is 0.7, and claims calculation agent task coreity is 0.8. Using a weighted algorithm (coreity weight 0.5, performance metric weight 0.5), priority weights are calculated: image recognition agent priority weight 0.76, repair quotation agent priority weight 0.48, and claims calculation agent priority weight 0.89. By combining dependencies (image recognition - repair quotation - claims calculation), the execution order is optimized: the image recognition agent is started first, and computing resources are pre-allocated to the high-weight claims calculation agent; after the repair quotation is completed, the claims calculation agent with pre-allocated resources is scheduled, which shortens the average time compared to the initial process sheet, forming the target orchestration process.

[0077] This application addresses the lack of scientific planning in process execution order in existing technologies through multi-dimensional data fusion and logical constraint optimization. First, it obtains operational performance indicators from the agent capability library to provide performance-dimensional data support for optimization. Then, using a preset weighting algorithm, it combines the task coreness reflecting business importance with operational performance indicators reflecting execution efficiency to calculate objective priority weights. Finally, it uses these priority weights as the scheduling basis while strictly adhering to the dependencies between agents to adjust the execution order, ensuring that the optimized target orchestration process both considers business priorities and guarantees logical coherence, thereby improving the overall efficiency of multi-agent collaboration.

[0078] In some optional implementations, after step 207, where the target orchestration process is executed to output the execution result of the target task, the following steps are also included: The execution results are parsed to obtain the task results and execution logs. The execution logs contain the dependencies between the agents and the conflict handling records. The task objectives, task results, and execution logs are stored in the training library so that the semantic analysis model, agent capability library, and conflict detector can be iterated based on the task objectives, task results, and execution logs in the training library.

[0079] The task results are the core output data directly related to the target task, generated after the execution of the target orchestration process. They must meet the output requirements in the task information and intuitively reflect the task completion status. The execution log is a full record document automatically generated during the execution of the target orchestration process. In addition to containing the dependencies of each agent, it also includes information such as the start / end time of each target agent, resource usage, and data flow nodes, providing a basis for tracing the details of process execution and troubleshooting.

[0080] Among them, the conflict handling record is a special record in the execution log regarding resource conflicts. It details the conflict detection results (such as conflicting agents and conflicting resources), conflict resolution strategies (such as time-sharing access and resource pre-allocation), and resolution effects (such as whether the conflict was eliminated and changes in execution time). It is key data for optimizing conflict handling logic. The training library is a structured database used to store task objectives, task results, and execution logs. By accumulating process execution data in multiple scenarios, it provides data support for subsequent iterations of the semantic analysis model (to improve task splitting accuracy), the agent capability library (to update performance metrics), and the conflict detector (to optimize resolution strategies).

[0081] In one example, in a medical insurance claims review scenario, the claims review result is output after the target orchestration process is executed. This result is parsed to obtain the task result and execution log. The task result is "the insured's medical insurance reimbursement amount is 8500 yuan, and the reimbursement ratio is 80%"; the execution log includes the agent dependencies "medical record extraction - medical insurance catalog matching - amount calculation" and conflict handling records "the conflict between medical record extraction and cost verification agent reading was resolved using a time-sharing access strategy, reducing processing time." The task target "medical insurance claims review," the task result, and the execution log are stored in the training library. Based on multiple batches of data accumulated in the training library, the semantic analysis model is iterated to improve its task splitting accuracy, the performance indicators of each agent in the agent capability library are updated, the strategy matching logic of the conflict detector is optimized, and the efficiency of subsequent collaborative processing of the system is improved.

[0082] This application addresses the lack of self-optimization capabilities in existing multi-agent collaborative systems by analyzing execution results, accumulating full data, and iterating on core components. First, the execution results are broken down to obtain the core task outcomes and execution logs containing dependencies and conflict resolution records, thus retaining complete data across the entire execution process. Then, the task objectives, results, and execution logs are stored in a training library to provide real-world training data for the semantic analysis model, agent capability library, and conflict detector. This enables continuous iterative optimization of system components, improving the accuracy of task breakdown, agent matching efficiency, and conflict resolution effectiveness.

[0083] It should be emphasized that, in order to further ensure the privacy and security of the above-mentioned task information, intelligent agent recommendation list, conflict resolution strategy, and execution results, the above-mentioned task information, intelligent agent recommendation list, conflict resolution strategy, and execution results can also be stored in a blockchain node.

[0084] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0085] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0086] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0087] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a task processing device based on an intelligent agent orchestration canvas. This device embodiment is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0088] like Figure 3 As shown, the task processing device 400 based on intelligent agent orchestration canvas in this embodiment includes: a receiving module 401, a splitting module 402, an acquisition module 403, a detection module 404, a generation module 405, a processing module 406, and an optimization module 407. Wherein: The receiving module 401 is used to receive the task information of the target task input by the user in the intelligent agent arrangement canvas; The splitting module 402 is used to split the task objective in the task information into a chain of sub-tasks by using the semantic analysis model built into the intelligent agent orchestration canvas. The acquisition module 403 is used to acquire the target intelligent agent corresponding to the sub-task chain from the preset intelligent agent capability library, generate an intelligent agent recommendation list, and construct a preliminary orchestration process based on the intelligent agent recommendation list; The detection module 404 is used to detect whether there are resource conflicts between the target intelligent agents in the preliminary arrangement process using a preset conflict judgment model. The generation module 405 is used to generate a conflict resolution strategy based on the execution constraints in the task information and the task core degree of each target agent if there is a resource conflict, by using a preset conflict detector. Processing module 406 is used to perform conflict resolution processing on the preliminary orchestration process based on the conflict resolution strategy to obtain an orchestration process without resource conflicts. The optimization module 407 is used to optimize the execution order of the orchestration process without resource conflicts based on the running performance indicators and task coreness of each target agent, to obtain the target orchestration process, execute the target orchestration process, and output the execution result of the target task.

[0089] This application's embodiments address the core shortcomings of existing multi-agent collaborative technologies through multi-stage collaborative design, significantly improving the efficiency and reliability of complex task execution. Leveraging the semantic analysis model built into the agent orchestration canvas, target tasks are automatically decomposed, transforming high-level tasks into logically coherent sub-task chains. This eliminates the need for manual decomposition, simplifying the process and ensuring the rationality of the decomposition, laying the foundation for collaborative execution. Through an agent capability library, target agents are automatically matched and adapted based on the sub-task chains to construct a preliminary orchestration process, accurately identifying agent execution characteristics and relationships, avoiding the subjectivity and inefficiency of manual matching. The collaboration of a conflict judgment model and a conflict detector enables automated detection and resolution of resource conflicts. Combined with task execution constraints and agent task coreity, customized strategies are generated, effectively avoiding data corruption and task interruptions. Simultaneously, the execution order is optimized based on agent performance indicators and task coreity, ensuring the scientific and efficient nature of the process and comprehensively improving the intelligence level and execution reliability of multi-agent collaborative processing of complex tasks.

[0090] In one embodiment, the acquisition module 403 includes: The calling submodule is used to call the intelligent agent capability library built into the intelligent agent orchestration canvas. The intelligent agent capability library stores the capability descriptions, data interaction specifications and access resource ranges of each candidate intelligent agent. The first generation submodule is used to generate the capability feature vector of each candidate agent based on the capability description, data interaction specifications and access resource scope. The second generation submodule is used to generate the task feature vector of each subtask in the subtask chain through a semantic analysis model; The first calculation submodule is used to calculate the matching degree between the capability feature vector and each task feature vector, select candidate agents with matching degrees greater than a preset matching degree threshold as target agents, and generate an agent recommendation list.

[0091] In one embodiment, the acquisition module 403 includes: The integration submodule is used to integrate task information with the capability descriptions, data interaction specifications, and access resource scope of each target intelligent agent in the intelligent agent recommendation list through a semantic analysis model, and generate capability embedding vectors for each target intelligent agent. The second computational submodule is used to calculate the input-output similarity between adjacent target agents in each target agent, and the resource-task similarity between each target agent and the target task, based on the capability embedding vector. A submodule is established to create dependency connections between target agents in the agent orchestration canvas and generate a preliminary orchestration process if the input-output similarity and resource-task similarity are both greater than a preset threshold.

[0092] In one embodiment, the second calculation submodule is further configured to: split each capability embedding vector into an input feature subvector, an output feature subvector, and a resource requirement subvector; calculate the similarity between the output feature subvector of the adjacent preceding target agent and the input feature subvector of the adjacent subsequent target agent for all target agents, thereby obtaining the input-output similarity between adjacent target agents; parse the task information to obtain the resource requirement feature vector of the target task; and calculate the similarity between the resource requirement subvector of each target agent and the resource requirement feature vector, thereby obtaining the resource task similarity between each target agent and the target task.

[0093] In one embodiment, the detection module 404 includes: The first extraction submodule is used to extract the access resource set of each target agent from the preliminary orchestration process. The access resource set includes the read resource set and the write resource set. The detection submodule is used to match and detect the resource access behavior of each target agent in the initial orchestration process based on the read resource set and the write resource set, and through a preset conflict judgment model, and outputs a conflict detection result dataset. The first determination submodule is used to determine, based on the conflict detection result dataset, that there is a resource conflict between the target agents in the preliminary orchestration process if there is a write conflict, a read conflict, or a read-write conflict between any two target agents. The second determination submodule is used to determine that there are no resource conflicts among the target agents in the initial orchestration process if there are no write conflicts, read conflicts, or read-write conflicts among all target agents.

[0094] In one embodiment, the generation module 405 includes: The second extraction submodule is used to extract the original execution sequence and conflict type of the preliminary orchestration process from the conflict detection result dataset; The third calculation submodule is used to calculate the task core degree of each target agent based on the relationship between the sub-task chain and the target task using a preset algorithm. The determination submodule is used to call the preset conflict resolution mapping library, and determine the conflict resolution strategy based on the original execution sequence, combined with the conflict type, execution constraints and task coreness.

[0095] In one embodiment, the optimization module 407 includes: The acquisition submodule is used to obtain the operational performance indicators of each target intelligent agent from the intelligent agent capability library; The fourth calculation submodule is used to calculate the priority weight of each target agent based on the task coreness and running performance indicators, according to a preset weighting algorithm. The optimization submodule is used to optimize the execution order of orchestration processes without resource conflicts based on priority weights and dependencies between target agents, and generate target orchestration processes.

[0096] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0097] Computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0098] Computer devices can include desktop computers, laptops, handheld computers, and cloud servers. These devices allow for human-computer interaction with users through keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0099] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and the external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for a task processing method based on an intelligent agent's canvas arrangement. In addition, memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0100] In some embodiments, processor 62 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. This processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, processor 62 is used to execute computer-readable instructions stored in memory 61 or to process data, such as executing computer-readable instructions for a task processing method based on an agent-based orchestration canvas.

[0101] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the computer device 6 and other electronic devices.

[0102] This application's embodiments address the core shortcomings of existing multi-agent collaborative technologies through multi-stage collaborative design, significantly improving the efficiency and reliability of complex task execution. Leveraging the semantic analysis model built into the agent orchestration canvas, target tasks are automatically decomposed, transforming high-level tasks into logically coherent sub-task chains. This eliminates the need for manual decomposition, simplifying the process and ensuring the rationality of the decomposition, laying the foundation for collaborative execution. Through an agent capability library, target agents are automatically matched and adapted based on the sub-task chains to construct a preliminary orchestration process, accurately identifying agent execution characteristics and relationships, avoiding the subjectivity and inefficiency of manual matching. The collaboration of a conflict judgment model and a conflict detector enables automated detection and resolution of resource conflicts. Combined with task execution constraints and agent task coreity, customized strategies are generated, effectively avoiding data corruption and task interruptions. Simultaneously, the execution order is optimized based on agent performance indicators and task coreity, ensuring the scientific and efficient nature of the process and comprehensively improving the intelligence level and execution reliability of multi-agent collaborative processing of complex tasks.

[0103] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the task processing method based on the agent-based orchestration canvas as described above.

[0104] This application's embodiments address the core shortcomings of existing multi-agent collaborative technologies through multi-stage collaborative design, significantly improving the efficiency and reliability of complex task execution. Leveraging the semantic analysis model built into the agent orchestration canvas, target tasks are automatically decomposed, transforming high-level tasks into logically coherent sub-task chains. This eliminates the need for manual decomposition, simplifying the process and ensuring the rationality of the decomposition, laying the foundation for collaborative execution. Through an agent capability library, target agents are automatically matched and adapted based on the sub-task chains to construct a preliminary orchestration process, accurately identifying agent execution characteristics and relationships, avoiding the subjectivity and inefficiency of manual matching. The collaboration of a conflict judgment model and a conflict detector enables automated detection and resolution of resource conflicts. Combined with task execution constraints and agent task coreity, customized strategies are generated, effectively avoiding data corruption and task interruptions. Simultaneously, the execution order is optimized based on agent performance indicators and task coreity, ensuring the scientific and efficient nature of the process and comprehensively improving the intelligence level and execution reliability of multi-agent collaborative processing of complex tasks.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0106] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

[0107] The software tools or components not belonging to our company that appear in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A task processing method based on intelligent agent-based canvas arrangement, characterized in that, Includes the following steps: Receive task information of the target task input by the user in the intelligent agent orchestration canvas; The semantic analysis model built into the intelligent agent orchestration canvas is used to break down the task objectives in the task information into sub-task chains. Obtain the target intelligent agent corresponding to the sub-task chain from the preset intelligent agent capability library, generate an intelligent agent recommendation list, and construct a preliminary orchestration process based on the intelligent agent recommendation list; A preset conflict judgment model is used to detect whether there are resource conflicts between the target agents in the preliminary orchestration process; If resource conflicts exist, a conflict resolution strategy is generated based on the execution constraints in the task information and the task core degree of each target agent through a preset conflict detector. Based on the conflict resolution strategy, the initial orchestration process is processed to resolve conflicts, resulting in an orchestration process without resource conflicts. Based on the operational performance indicators of each target agent and the task coreness, the execution order of the resource-free orchestration process is optimized to obtain the target orchestration process. The target orchestration process is then executed to output the execution result of the target task.

2. The method according to claim 1, characterized in that, The step of obtaining the target intelligent agent corresponding to the sub-task chain from the preset intelligent agent capability library and generating an intelligent agent recommendation list specifically includes: The system invokes the built-in agent capability library of the agent orchestration canvas, which stores the capability descriptions, data interaction specifications, and access resource ranges of each candidate agent. Based on the capability description, the data interaction specifications, and the access resource range, a capability feature vector is generated for each candidate agent. The semantic analysis model is used to generate a task feature vector for each subtask in the subtask chain. Calculate the matching degree between the capability feature vector and each task feature vector, select candidate agents with matching degrees greater than a preset matching degree threshold as target agents, and generate an agent recommendation list.

3. The method according to claim 1, characterized in that, The step of constructing a preliminary orchestration process based on the agent recommendation list specifically includes: The semantic analysis model is used to integrate the task information with the capability descriptions, data interaction specifications, and access resource ranges of each target intelligent agent in the intelligent agent recommendation list to generate capability embedding vectors for each target intelligent agent. Based on the capability embedding vector, the input-output similarity between adjacent target agents in each target agent and the resource-task similarity between each target agent and the target task are calculated respectively. If the input-output similarity and the resource-task similarity are both greater than a preset threshold, then dependency connections between the target agents are established in the agent orchestration canvas to generate a preliminary orchestration process.

4. The method according to claim 3, characterized in that, The steps of calculating the input-output similarity between adjacent target agents and the resource-task similarity between each target agent and the target task based on the capability embedding vector specifically include: For each capability embedding vector, it is split into an input feature sub-vector, an output feature sub-vector, and a resource requirement sub-vector; For all target agents, calculate the similarity between the output feature vector of the adjacent preceding target agent and the input feature vector of the adjacent subsequent target agent to obtain the input-output similarity between adjacent target agents in each target agent; The task information is parsed to obtain the resource requirement feature vector of the target task; Calculate the similarity between the resource requirement subvector and the resource requirement feature vector of each target agent to obtain the resource task similarity between each target agent and the target task.

5. The method according to claim 1, characterized in that, The step of using a preset conflict judgment model to detect whether there are resource conflicts among the target agents in the preliminary orchestration process specifically includes: The access resource set of each target agent is extracted from the preliminary orchestration process, and the access resource set includes a read resource set and a write resource set. Based on the read resource set and the write resource set, a preset conflict judgment model is used to match and detect the resource access behavior of each target agent in the preliminary orchestration process, and output a conflict detection result dataset. Based on the conflict detection result dataset, if there is a write conflict, a read conflict, or a read-write conflict between any two target agents, then it is determined that there is a resource conflict between the target agents in the preliminary orchestration process. If there are no write conflicts, read conflicts, or read-write conflicts among all target agents, then it is determined that there are no resource conflicts among the target agents in the preliminary orchestration process.

6. The method according to claim 5, characterized in that, The step of generating a conflict resolution strategy based on the execution constraints in the task information and the task coreness of each target agent using a preset conflict detector specifically includes: Extract the original execution sequence and conflict type of the preliminary orchestration process from the conflict detection result dataset; Using a preset algorithm, the task core degree of each target agent is calculated based on the association between the sub-task chain and the target task; A preset conflict resolution mapping library is invoked, and a conflict resolution strategy is determined based on the original execution sequence, the conflict type, the execution constraints, and the task coreity.

7. The method according to claim 1, characterized in that, The steps for optimizing the execution order of the resource-free orchestration process based on the operational performance indicators of each target agent and the task core degree to obtain the target orchestration process specifically include: Obtain the operational performance indicators of each target intelligent agent from the intelligent agent capability library; Based on the task coreness and the operational performance indicators, the priority weight of each target agent is calculated according to a preset weighting algorithm. Based on the priority weights and the dependencies between the target agents, the execution order of the orchestration process without resource conflicts is optimized to generate the target orchestration process.

8. A task processing device based on an intelligent agent arranging a canvas, characterized in that, include: The receiving module is used to receive task information of the target task input by the user in the intelligent agent arrangement canvas; The splitting module is used to split the task objective in the task information into a chain of sub-tasks using the semantic analysis model built into the intelligent agent's orchestration canvas. The acquisition module is used to acquire the target intelligent agent corresponding to the sub-task chain from the preset intelligent agent capability library, generate an intelligent agent recommendation list, and construct a preliminary orchestration process based on the intelligent agent recommendation list; The detection module is used to detect whether there are resource conflicts between the target agents in the preliminary orchestration process using a preset conflict judgment model. The generation module is used to generate a conflict resolution strategy based on the execution constraints in the task information and the task core degree of each target agent if resource conflicts exist, by using a preset conflict detector. The processing module is used to perform conflict resolution processing on the preliminary orchestration process based on the conflict resolution strategy to obtain an orchestration process without resource conflicts. The optimization module is used to optimize the execution order of the orchestration process without resource conflicts based on the running performance indicators of each target agent and the task coreness, to obtain the target orchestration process, execute the target orchestration process, and output the execution result of the target task.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the task processing method based on an agent-based canvas arrangement as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the task processing method based on an agent-based canvas arrangement as described in any one of claims 1 to 7.

Images