Search

CN-121998585-A - Task processing method, device, equipment and medium based on agent canvas

CN121998585ACN 121998585 ACN121998585 ACN 121998585ACN-121998585-A

Abstract

The application belongs to the field of artificial intelligence, and relates to a task processing method, device, equipment and medium based on an agent arranging canvas, which comprises the steps of receiving task information of a target task input by a user in the agent arranging canvas, and splitting the task target in the task information into a subtask chain by using a semantic analysis model built in the canvas; the method comprises the steps of obtaining corresponding target intelligent agents from an intelligent agent capability library, generating a recommendation list, constructing a preliminary arranging process, detecting resource conflict by using a conflict judging model, generating a resolution strategy by a conflict detector based on execution constraint and task core degree if the resource conflict exists, processing the conflict to obtain a conflict-free process, optimizing the execution sequence of the conflict-free process according to the running performance index of each intelligent agent and the task core degree to obtain the target arranging process, and executing the process to output a target task execution result. The application can be applied to the business fields of finance, science, insurance, medical treatment and the like, and can improve the efficiency of complex tasks executed by multi-agent cooperation and the reliability of task execution.

Inventors

  • LIN YANYU
  • TONG YAOYI
  • XU WEI

Assignees

  • 平安科技（深圳）有限公司

Dates

Publication Date
20260508
Application Date
20260109

Claims (10)

  1. 1. A task processing method based on an agent canvas is characterized by comprising the following steps: receiving task information of a target task input in an agent programming canvas by a user; splitting a task target in the task information into a subtask chain through a semantic analysis model built in the agent programming canvas; acquiring a target agent corresponding to the subtask chain from a preset agent capability library, generating an agent recommendation list, and constructing a preliminary arrangement flow according to the agent recommendation list; detecting whether resource conflict exists among all target intelligent agents in the preliminary arrangement flow by adopting a preset conflict judging model; If resource conflict exists, a conflict resolution strategy is generated based on the execution constraint in the task information and the task core degree of each target intelligent agent through a preset conflict detector; based on the conflict resolution strategy, carrying out conflict resolution processing on the preliminary arrangement flow to obtain an arrangement flow without resource conflict; And optimizing the execution sequence of the scheduling process without resource conflict based on the operation performance index of each target agent and the task core degree to obtain a target scheduling process, and executing the target scheduling process to output the execution result of the target task.
  2. 2. The method of claim 1, wherein the step of obtaining the target agent corresponding to the subtask chain from a preset agent capability library and generating an agent recommendation list specifically includes: Invoking an intelligent capability library built in the intelligent agent arrangement canvas, wherein the intelligent capability library stores capability descriptions, data interaction specifications and access resource ranges of candidate intelligent agents; Generating capability feature vectors of each candidate agent based on the capability description, the data interaction specification and the access resource range; Generating a task feature vector of each subtask in the subtask chain through the semantic analysis model; and calculating the matching degree between the capability feature vector and each task feature vector, and taking the candidate agent with the matching degree larger than a preset matching degree threshold as a target agent to generate an agent recommendation list.
  3. 3. The method according to claim 1, wherein the step of constructing a preliminary orchestration flow from the agent recommendation list specifically comprises: Integrating the task information with the capability description, the data interaction specification and the access resource range of each target intelligent agent in the intelligent agent recommendation list through the semantic analysis model to generate a capability embedding vector of each target intelligent agent; Based on the capability embedding vector, respectively calculating the input/output similarity between adjacent target intelligent agents in each target intelligent agent and the resource task similarity between each target intelligent agent and the target task; And if the input-output similarity and the resource task similarity are both larger than a preset threshold, establishing a dependent connection between the target agents in the agent arrangement canvas to generate a preliminary arrangement flow.
  4. 4. The method according to claim 3, wherein the step of calculating, based on the capability embedding vector, input/output similarities between adjacent target agents in the target agents, and resource task similarities between the target agents and the target tasks, respectively, specifically includes: Aiming at each capability embedding vector, splitting the capability embedding vector into an input characteristic sub-vector, an output characteristic sub-vector and a resource requirement sub-vector; Calculating the similarity between the output characteristic sub-vector of the adjacent preceding target agent and the input characteristic sub-vector of the adjacent following target agent aiming at all target agents to obtain the input and output similarity between the adjacent target agents in each target agent; Analyzing the task information to obtain a resource demand feature vector of the target task; And calculating the similarity between the resource demand sub-vector and the resource demand feature vector of each target agent to obtain the similarity of the resource task between each target agent and the target task.
  5. 5. The method according to claim 1, wherein the step of detecting whether resource conflicts exist among the target agents in the preliminary orchestration process by using a preset conflict judgment model specifically includes: extracting access resource sets of all target agents from the preliminary arrangement flow, wherein the access resource sets comprise a read resource set and a write resource set; Based on the read resource set and the write resource set, performing matching detection on the resource access behaviors of the target agents in the preliminary arrangement flow through a preset conflict judgment model, and outputting a conflict detection result data set; Based on the conflict detection result data set, if write-in conflict or read-out conflict or read-write conflict exists between any two target agents, judging that resource conflict exists between the target agents in the preliminary arrangement flow; And if the write conflict, the read conflict and the read-write conflict do not exist among all the target agents, judging that the resource conflict does not exist among all the target agents in the preliminary arrangement flow.
  6. 6. The method according to claim 5, wherein the step of generating, by a preset conflict detector, a conflict resolution policy based on the execution constraint in the task information and the task core degree of each target agent specifically includes: extracting the original execution time sequence and conflict type of the preliminary arrangement flow from the conflict detection result data set; calculating the task core degree of each target intelligent agent based on the association relation between the subtask chain and the target task through a preset algorithm; and calling a preset conflict disposal mapping library, and determining a conflict resolution strategy by taking the original execution time sequence as a reference and combining the conflict type, the execution constraint and the task core degree.
  7. 7. The method according to claim 1, wherein the step of optimizing the execution sequence of the resource conflict-free scheduling process based on the operation performance index of each target agent and the task core degree to obtain a target scheduling process specifically comprises: Acquiring the operation performance index of each target intelligent agent from the intelligent agent capability library; calculating the priority weight of each target intelligent agent according to a preset weight algorithm based on the task core degree and the running performance index; And optimizing the execution sequence of the scheduling process without resource conflict based on the priority weight and the dependency relationship among the target agents, and generating a target scheduling process.
  8. 8. A task processing device based on an agent orchestration canvas, comprising: the receiving module is used for receiving task information of a target task input in the agent arrangement canvas by a user; The splitting module is used for splitting the task target in the task information into a subtask chain through a semantic analysis model built in the agent programming canvas; The acquisition module is used for acquiring target agents corresponding to the subtask chain from a preset agent capability library, generating an agent recommendation list and constructing a preliminary arrangement flow according to the agent recommendation list; the detection module is used for detecting whether resource conflict exists among all target agents in the preliminary arrangement flow by adopting a preset conflict judgment model; the generation module is used for generating a conflict resolution strategy based on the execution constraint in the task information and the task core degree of each target intelligent agent through a preset conflict detector if resource conflict exists; The processing module is used for carrying out conflict resolution processing on the preliminary arrangement flow based on the conflict resolution strategy to obtain an arrangement flow without resource conflict; And the optimizing module is used for optimizing the execution sequence of the scheduling process without resource conflict based on the running performance index of each target agent and the task core degree to obtain a target scheduling process, and executing the target scheduling process to output the execution result of the target task.
  9. 9. A computer device comprising a memory having stored therein computer readable instructions which when executed by the processor implement the steps of the agent orchestration canvas based task handling method according to any one of claims 1 to 7.
  10. 10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor perform the steps of the agent orchestration canvas based task handling method according to any one of claims 1 to 7.

Description

Task processing method, device, equipment and medium based on agent canvas Technical Field The application relates to the technical field of artificial intelligence, is applied to online processing of business scenes of finance technology, insurance, medical treatment and the like, in particular to a task processing method, device, equipment and medium based on an agent canvas. Background Along with the large-scale landing of the large-model technology in an enterprise-level scene, multi-agent cooperation becomes a core mode for processing complex business tasks, and specialized agents such as data extraction, report generation, risk assessment and the like realize the efficient promotion of business processes through division cooperation. In a financial business scene, multi-agent cooperation is widely applied to key business links such as vehicle insurance damage assessment and settlement, medical insurance verification, credit risk assessment, intelligent consultation strategy generation and the like, the requirements of enterprises on multi-agent cooperation are increasingly urgent, but a plurality of short boards still exist in the current related technical scheme, and the intelligent and accurate requirements in practical application are difficult to meet. In the prior art, multi-agent cooperation lacks deep semantic parsing capability for tasks, and cannot automatically disassemble high-level target tasks proposed by users into logically coherent sub-task sequences, and manual splitting and configuration are needed, so that the operation is complex, and the overall execution effect is easily affected due to unreasonable splitting logic. Meanwhile, the system is difficult to accurately identify the execution characteristics and association relations of different agents, cannot automatically match the agents adapting to the sub-tasks, and cannot pre-judge the input and output dependence among the agents, so that the collaborative process lacks scientific execution sequence planning. In addition, for the possible resource conflicts of the same data source access, the same target field writing and the like of multiple agents, the existing scheme lacks an effective automatic detection and resolution mechanism, manual intervention is often needed for investigation, and the problems of low efficiency, data disorder, task execution interruption and the like are caused. In a financial business scene, the technical defects can directly lead to the problems of prolonged vehicle insurance claim aging, reduced verification accuracy of the insurance, increased credit risk assessment deviation and the like, so that the efficiency is low and the task execution reliability is poor when a plurality of agents cooperatively execute complex tasks. Disclosure of Invention The embodiment of the application aims to provide a task processing method, device, computer equipment and storage medium based on an agent canvas, which are used for solving the problems of low efficiency and poor task execution reliability when the traditional multi-agent cooperation executes complex tasks. In a first aspect, a task processing method based on an agent canvas is provided, which adopts the following technical scheme: The method comprises the steps of receiving task information of target tasks input by a user in an agent programming canvas, splitting task targets in the task information into subtask chains through a semantic analysis model built in the agent programming canvas, obtaining target agents corresponding to the subtask chains from a preset agent capability library, generating an agent recommendation list, constructing a preliminary programming flow according to the agent recommendation list, detecting whether resource conflicts exist among the target agents in the preliminary programming flow by adopting a preset conflict judgment model, generating a conflict resolution strategy based on execution constraints in the task information and task core degrees of the target agents through a preset conflict detector if the resource conflicts exist, carrying out conflict resolution on the preliminary programming flow based on the conflict resolution strategy to obtain a programming flow without resource conflicts, optimizing the execution sequence of the programming flow without resource conflicts based on the operation performance indexes and the task core degrees of the target agents to obtain the target programming flow, and executing the target programming flow to output execution results of the target tasks. In a second aspect, a task processing device based on an agent canvas is provided, which adopts the following technical scheme: the receiving module is used for receiving task information of a target task input in the agent arrangement canvas by a user; the splitting module is used for splitting the task targets in the task information into subtask chains through a semantic analysis model built in the agent programming canvas;