Computer Integrated Manufacturing System ›› 2026, Vol. 32 ›› Issue (3): 999-1010.DOI: 10.13196/j.cims.2025.BPM15

Previous Articles     Next Articles

Deterministic behavior recognition method based on control and data flow

XU Wenru,FANG Xinyang,YU Dongjin+,SUN Xiaoxiao,CHENG Yuying   

  1. School of Computer Science and Technology,Hangzhou Dianzi University
  • Online:2026-03-31 Published:2026-04-03
  • Supported by:
    Project supported by the Zhejiang Provincial Natural Science Foundation,China(No.LZ25F020011),and  the National Natural Science Foundation,China(No.62372145).

基于控制流和数据流的确定性行为识别方法

徐闻如,方鑫杨,俞东进+,孙笑笑,程玉莹   

  1. 杭州电子科技大学计算机学院
  • 作者简介:
    徐闻如(2000-),男,江苏苏州人,硕士研究生,研究方向:业务流程管理,E-mail:222050138@hdu.edu.cn;

    方鑫杨(1998-),男,浙江衢州人,硕士研究生,研究方向:业务流程管理,E-mail:fangxy926@hdu.edu.cn;

    +俞东进(1969-),男,浙江杭州人,教授,博士,博士生导师,研究方向:业务流程管理、服务计算、大数据与软件工程等,通讯作者,E-mail:yudj@hdu.edu.cn;

    孙笑笑(1991-),女,浙江浦江人,副教授,博士,硕士生导师,研究方向:业务流程管理、时空数据分析等,E-mail:sunxiaoxiao@hdu.edu.cn;

    程玉莹(2001-),女,山东济南人,硕士研究生,研究方向:业务流程管理,E-mail:242050130@hdu.edu.cn。
  • 基金资助:
    浙江省自然科学基金重点资助项目(LZ25F020011);国家自然科学基金资助项目(62372145)。

Abstract: Deterministic behavior recognition is one of the important tasks in robotic process mining research and the basis for automating robotic processes.To addresses the problem,most existing methods identify only part of the deterministic behavior from user interaction logs,which leads to incomplete identification of deterministic behavior and incorrect execution of automated programs.Therefore,a method for identifying deterministic actions based on control-flow and data-flow perspectives was proposed.It first utilized process discovery algorithms to obtain the process model from task traces,and the determinacy of the type and parameters of actions from both control-flow and data-flow perspectives were analyzed.By considering the impact of prefix trace action's data attributes and prefix trace patterns on control-flow decision making,the control-flow perspective used machine learning models to predict the type and location parameters of the next action.Meanwhile,the data-flow perspective analyzed the influence of data transformation dependency between actions and the influence of prefix trajectory behavior type on data attribute.In this perspective,data transformation rules were mined,and multiple predictive models were trained to judge the data parameter of the next action.Finally,based on the results of the control-flow and data-flow perspectives,a final judgement on deterministic actions was formed.The effectiveness and interpretability of the proposed method was verified by taking the example of a real company's login,outbound and purchase business processes.

Key words: robotic process automation, robotic process mining, user interaction logs, deterministic action identification

摘要: 确定性行为识别是机器人流程挖掘研究中的重要任务之一,也是实现机器人流程自动化的基础。大多数现有方法仅从用户交互日志中识别部分确定性行为,导致识别不全面,进而影响自动化程序的正确执行。为解决该问题,提出了一种基于控制流和数据流的确定性行为识别方法。该方法首先利用流程发现算法从任务轨迹中获取行为流程模型,然后分别从控制流和数据流的角度,分析行为类型与属性的确定性。其中,控制流角度考虑了前缀轨迹行为的数据属性和前缀轨迹模式对流程决策的影响,通过训练机器学习模型来预测下一个行为的类型和位置属性;数据流角度考虑了行为间的数据转换关系以及前缀轨迹行为类型对数据属性的影响,通过挖掘数据转换规则和训练多个预测模型来确定下一个行为的数据属性。最后,结合控制流和数据流角度的识别结果,对确定性行为做出最终判定。本文以某真实公司登录、出库采购业务流程为例,验证了所提方法的有效性和可解释性。

关键词: 机器人流程自动化, 机器人流程挖掘, 用户交互日志, 确定性行为识别

CLC Number: