CN-122022433-A - Workflow-based agent implementation method and system

CN122022433ACN 122022433 ACN122022433 ACN 122022433ACN-122022433-A

Abstract

The application discloses an agent realization method and system based on workflow, and relates to the technical field of agent workflow. The method comprises the steps of obtaining a flow of a work task, dividing the flow of the work task into a plurality of nodes to form a plurality of basic workflows, obtaining historical work task information of a user, evaluating to obtain a basic confidence value of the user on the work task, obtaining basic work nodes in the basic workflows, evaluating to obtain node confidence values of the basic work nodes in combination with the basic confidence values, obtaining the basic success rate of the basic work nodes in combination with the node confidence values to obtain the node success rate of the basic work nodes, evaluating to obtain the workflow success rate of the basic workflows, evaluating the workflow loss stopping capacity value of the user on the basic workflows, selecting to obtain user workflows in combination with the workflow success rate, and pushing the user workflows to a user side. The application improves the dynamic adaptability of the intelligent agent realization based on the workflow.

Inventors

LIN JIAXIN
Huang Hanye
HUANG GUANGJING
HUANG YUETIAN
LI RUIQI
YANG YONGJIAO
YAN YUPING
HUANG SHUWEI
ZHU ZEQI
LU QING
QIN QIANG

Assignees

广东电网有限责任公司
广东电网有限责任公司数智运营中心

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The method for realizing the intelligent agent based on the workflow is characterized by comprising the following steps: The method comprises the steps of obtaining a flow of a work task, dividing the flow of the work task into a plurality of nodes, and forming a plurality of basic workflows according to node arrangement and combination; acquiring historical work task information of a user, and evaluating according to the historical work task information to obtain a basic confidence value of the user on a work task; Acquiring a basic working node in a basic workflow, marking a node in front of the basic working node as a front node, and obtaining a node confidence value of the basic working node by combining basic confidence value evaluation; Obtaining a basic success rate of the basic working node, obtaining the node success rate by combining the node confidence value with the basic working node, and obtaining a workflow success rate of the basic workflow according to the node success rate evaluation; And evaluating the workflow loss stopping capability value of the user on the basic workflow according to the historical work task information, selecting the user workflow from the basic workflow in combination with the workflow success rate, and pushing the user workflow to the user side.
2. The method for realizing the agent based on the workflow according to claim 1, wherein the step of obtaining the historical work task information of the user and obtaining the basic confidence value of the user to the work task according to the historical work task information is specifically as follows: Judging whether the user executes the same work task according to the historical work task information, and if so, extracting the same work task information from the historical work task information; acquiring average execution time length of all users in the work task, counting the average time length of the users executing the work task, recording the average time length as the user execution time length, and calculating to obtain a time length difference value between the average execution time length and the user execution time length; The method comprises the steps of combining the same work task information and a time length difference value to obtain a proficiency confidence value of a user, wherein the same work task information comprises a historical task success rate and a historical task average execution frequency; if the same work task is not executed, acquiring a similar work task, and acquiring a proficiency confidence value according to the work evaluation of the similar task; And acquiring the real-time state of the user, and evaluating the state confidence value according to the real-time state of the user, wherein the state confidence value is combined with the basic confidence value of the user on the work task.
3. The method for realizing the agent based on the workflow according to claim 2, wherein if the same task is not executed, the step of obtaining a similar task and obtaining a proficiency confidence value according to the task work evaluation is specifically as follows: extracting historical tasks according to the historical work task information, comparing task similarity of the historical tasks and the work tasks, and screening according to the task similarity to obtain similar work tasks; historical task information of similar work tasks is obtained, an average value of task similarity is calculated, and a similar confidence value is obtained by combining the historical task information; extracting task thinking modes of historical tasks, comparing the thinking similarity of the task thinking modes of the historical tasks and the task thinking modes of the work tasks, and screening to obtain the thinking work tasks according to the thinking similarity; evaluating the knowledge migration capability value of the user, calculating the average value of the thinking similarity of the thinking work task, and calculating to obtain a thinking confidence value by combining the knowledge migration capability value; and combining the similar confidence value and the thinking confidence value to comprehensively obtain a proficiency confidence value.
4. A workflow-based agent implementation method according to claim 3, characterized by the step of evaluating the user's knowledge migration capability value, in particular: Counting the number of tasks of the same thinking mode between the same thinking mode actively applied by the user and the first application and recording the number as the active execution number; Counting the average time length between the same thinking mode of the active application of the user and the first application and recording the average time length as the active execution time length; counting the number of task fields contained in historical tasks corresponding to the same thinking mode, and counting the correct application rate of migrating the same thinking mode; And obtaining the knowledge migration capability value of the user by combining the number of active executions, the active execution duration, the number of task fields and the correct application rate.
5. The method for realizing the intelligent agent based on the workflow according to claim 2, wherein the step of obtaining the real-time state of the user and obtaining the state confidence value according to the real-time state evaluation of the user is specifically as follows: Extracting according to the historical work task information to obtain a user standard state, and comparing to obtain the state similarity of the user real-time state and the user standard state; Collecting a real-time task scene of a work task, and counting the work presentation difficulty of the real-time task scene; and obtaining the fatigue degree of the user, and comprehensively obtaining a state confidence value by combining the state similarity and the work presentation difficulty.
6. The method for realizing the agent based on the workflow according to claim 1, wherein the step of obtaining the basic working node in the basic workflow, marking the node in front of the basic working node as a front node, and obtaining the node confidence value of the basic working node by combining the basic confidence value evaluation comprises the following steps: Acquiring node completion difficulty of a basic working node, and counting association relation between the node completion difficulty and a confidence value; searching and obtaining a confidence value corresponding to the front node according to the association relation, and summing to obtain a front confidence value; And counting the number of the prepositions, and combining the prepositions with the node confidence values of the basic working nodes.
7. The method for realizing the intelligent agent based on the workflow according to claim 1, wherein the step of obtaining the success rate of the workflow of the basic workflow according to the node success rate evaluation by obtaining the success rate of the basic workflow by combining the node confidence value with the basic workflow is specifically as follows: Setting a node success rate threshold, and taking a basic working node with the node success rate lower than the node success rate threshold as a failure node; Counting the number of the nodes of the failed node, calculating the average value of the node success rate of the failed node and recording the average value as the average success rate; and obtaining the average influence degree of the failed nodes on the work task, and obtaining the workflow success rate of the basic workflow by combining the number of the nodes and the average success rate evaluation.
8. The method for realizing the intelligent agent based on the workflow according to claim 7, wherein the step of evaluating the workflow damage stopping capability value of the user to the basic workflow according to the historical work task information comprises the following specific steps: Acquiring processing information of a user on a historical work task, and evaluating according to the processing information to obtain a loss stopping capability value of the user; Acquiring average success rates of all users of the failed node, recording the average success rates as other success rates, and acquiring the task amount of the front node to the task amount ratio of the basic workflow; acquiring the prepositive failure number of the failure nodes in the prepositive nodes, acquiring the number of the failure nodes after the basic working nodes and recording the number as the postposition failure number; calculating the difference value between the number of post failures and the number of pre failures, and combining other success rates and the task quantity duty ratio to obtain a node loss stopping value of the basic working node; Combining the user loss stopping capability value and the node loss stopping value, and the workflow loss stopping capability value of the user on the basic workflow.
9. The method for realizing the intelligent agent based on the workflow as set forth in claim 8, wherein the step of obtaining the processing information of the user on the historical work task and obtaining the loss stopping ability value of the user according to the evaluation of the processing information comprises the following steps: searching a failure task of a user by using the processing information, and collecting the average duration of failure nodes of the user in the failure task; Recording the failed task with the average duration reaching the preset adherence threshold value as a damaged task, otherwise, recording the failed task as a non-damaged task; counting the average execution node number of the loss stopping task and recording the number of the loss stopping nodes; the average duration of the non-loss stopping task at the failed node is counted and recorded as the non-loss stopping duration; And combining the number of loss stopping nodes and the non-loss stopping time length to comprehensively obtain the loss stopping capability value of the user.
10. A workflow-based agent implementation system, wherein a workflow-based agent implementation method according to any one of claims 1-9 is applied, comprising: The basic sequence module is used for acquiring the flow of the work task, dividing the flow of the work task into a plurality of nodes, and forming a plurality of basic workflows according to node arrangement and combination; The basic confidence module is used for acquiring historical work task information of the user and obtaining a basic confidence value of the user on the work task according to the evaluation of the historical work task information; The node confidence module is used for acquiring basic working nodes in the basic workflow, marking the nodes in front of the basic working nodes as front nodes, and obtaining the node confidence value of the basic working nodes by combining with the basic confidence value evaluation; The work success module is used for acquiring the basic success rate of the basic work node, obtaining the node success rate by combining the node confidence value with the basic work node, and obtaining the workflow success rate of the basic workflow according to the node success rate evaluation; And the loss stopping selection module evaluates the workflow loss stopping capability value of the user on the basic workflow according to the historical work task information, and combines the workflow success rate to select the user workflow from the basic workflow and push the user workflow to the user side.

Description

Workflow-based agent implementation method and system Technical Field The application relates to the technical field of workflow of an agent, in particular to a workflow-based agent implementation method and a workflow-based agent implementation system. Background In the current development of workflow-based intelligent system technology, existing methods mainly aim at improving the automation degree and reliability of task execution through predefined flow rules and static node logic. However, the existing systems generally do not introduce an effective dynamic policy controller or a real-time state-aware mechanism, and cannot perform runtime optimization or node-level adaptive reconstruction on the workflow. In addition, the existing scheme depends on manual configuration and debugging, so that not only is the technical threshold and development cost increased, but also the expandability of the workflow and the flexibility of autonomous adjustment of a user are limited. This stiff design model makes it difficult for workflow agents to achieve true "intelligent adaptation" in open environments or cross-domain services, ultimately resulting in reduced user experience and inefficient execution. Disclosure of Invention The invention aims to provide a workflow-based agent implementation method and a workflow-based agent implementation system, which are used for solving the problems in the background technology. In a first aspect, the present application provides a workflow-based agent implementation method, which adopts the following technical scheme: The method comprises the steps of obtaining a flow of a work task, dividing the flow of the work task into a plurality of nodes, and forming a plurality of basic workflows according to node arrangement and combination; acquiring historical work task information of a user, and evaluating according to the historical work task information to obtain a basic confidence value of the user on a work task; Acquiring a basic working node in a basic workflow, marking a node in front of the basic working node as a front node, and obtaining a node confidence value of the basic working node by combining basic confidence value evaluation; Obtaining a basic success rate of the basic working node, obtaining the node success rate by combining the node confidence value with the basic working node, and obtaining a workflow success rate of the basic workflow according to the node success rate evaluation; And evaluating the workflow loss stopping capability value of the user on the basic workflow according to the historical work task information, selecting the user workflow from the basic workflow in combination with the workflow success rate, and pushing the user workflow to the user side. Preferably, the step of obtaining the historical work task information of the user and evaluating and obtaining the basic confidence value of the user to the work task according to the historical work task information comprises the following specific steps: Judging whether the user executes the same work task according to the historical work task information, and if so, extracting the same work task information from the historical work task information; acquiring average execution time length of all users in the work task, counting the average time length of the users executing the work task, recording the average time length as the user execution time length, and calculating to obtain a time length difference value between the average execution time length and the user execution time length; The method comprises the steps of combining the same work task information and a time length difference value to obtain a proficiency confidence value of a user, wherein the same work task information comprises a historical task success rate and a historical task average execution frequency; if the same work task is not executed, acquiring a similar work task, and acquiring a proficiency confidence value according to the work evaluation of the similar task; And acquiring the real-time state of the user, and evaluating the state confidence value according to the real-time state of the user, wherein the state confidence value is combined with the basic confidence value of the user on the work task. Preferably, if the same task is not executed, a similar task is obtained, and a skilled confidence value is obtained according to the task evaluation of the similar task, specifically: extracting historical tasks according to the historical work task information, comparing task similarity of the historical tasks and the work tasks, and screening according to the task similarity to obtain similar work tasks; historical task information of similar work tasks is obtained, an average value of task similarity is calculated, and a similar confidence value is obtained by combining the historical task information; extracting task thinking modes of historical tasks, comparing the thinking similarity of the task thinking modes