基于TAC-A2C的规范性业务流程监控方法  

Two agent cooperative-advanced actor critic method for prescriptive process monitoring

作　　者：朱开心 卢可 ZHU Kaixin;LU Ke(School of Mathematics&Big Data,Anhui University of Science&Technology,Huainan 232001,China;Anhui Province Engineering Laboratory for Big Data Analysis&Early Warning Technology of Coal Mine Safety,Huainan 232001,China)

机构地区：[1]安徽理工大学数学与大数据学院,安徽淮南232001 [2]安徽省煤矿安全大数据分析与预警技术工程实验室,安徽淮南232001

出　　处：《哈尔滨商业大学学报(自然科学版)》2025年第1期113-119,共7页Journal of Harbin University of Commerce:Natural Sciences Edition

基　　金：国家自然科学基金资助项目(61572035)。

摘　　要：规范性业务流程监控用于实时监测业务执行过程,及时发现并纠正偏离流程的异常行为.当前研究主要集中在使用统计分析、人工规则或深度学习技术来检测流程中的异常行为,并发出警报.对于异常检测之后如何进行流程调整以及如何对执行干预措施后的效果进行评估的问题,目前还缺乏行之有效的方法.针对该问题,提出了基于双智能体合作关系的Actor Critic(TAC-A2C)强化学习算法,该方法旨在通过两个智能体的协作来优化规范性业务流程监控任务.其中异常检测智能体负责训练策略网络以辨识流程中的潜在异常行为,并触发警报;流程调整智能体则负责训练策略网络来选择合适的流程干预行为,并评估其效果.通过训练中央控制器的价值网络,实现两个智能体之间的有效信息同步,从而实现紧密的合作.通过在事件日志上的实验评估表明,TAC-A2C方法能够有效检测到流程中的异常,并及时采取措施进行修复.Prescriptive process monitoring was used to track the execution of business processes in real time,promptly detecting and correcting abnormal behavior that deviated from the processes.Current research primarily focused on using statistical analysis,Heuristic rules or deep learning to detect anomalies in the process flow and issue alerts.However,there was a lack of effective methods for determining what actions to take after an anomaly was detected,how to adjust the process,and how to evaluate the effectiveness of the intervention measures.In response to this issue,this paper proposed a method called two agent cooperative-advanced actor critic(TAC-A2C),which aimed to optimize the prescriptive process monitoring task.The anomaly detection agent was responsible for training a policy network to identify potential abnormal behaviors in the process flow and trigger alerts.The process adjustment agent was responsible for training a policy network to select appropriate process intervention actions and assess their effectiveness.At the same time,by training the value network of a central controller,information synchronization between the two agents was achieved,enabling them to work closely together.Through experiments on event logs,it was demonstrated that the TAC-A2C method could effectively detect anomalies in the process and take timely measures to repair the process.

关 键 词：规范性业务流程监控 业务流程管理 异常检测 流程干预 深度强化学习 双智能体 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]