融合记忆网络与变迁系统的业务过程实例剩余执行时间预测方法
Business process instance remaining execution time prediction by integrating memory network and transition system
倪维健 1马少军 1刘彤 1曾庆田 1闫鸣2
作者信息
- 1. 山东科技大学计算机科学与工程学院,山东 青岛 266510
- 2. 济宁医学院附属医院,山东 济宁 272007
- 折叠
摘要
在业务过程实例剩余执行时间预测任务中,大多数现有方法仅对过程实例前缀进行编码,这类预测方法仅可以考虑到已经发生的事件对于预测结果的影响,在一定程度上忽略了过程实例中未来可能发生的事件对剩余时间预测的影响.针对这种不足,提出一种融合记忆网络与变迁系统的过程实例剩余时间预测方法.首先,通过变迁系统挖掘过程实例后缀,以双向门控循环单元为基础,对过程实例前缀和实例后缀进行编码;之后,使用记忆网络结合注意力机制学习过程实例中不同事件的权重,以便更好地对过程实例建模.此外,为充分考虑不同长度的过程实例前缀之间的相关性和数量差异,引入迁移学习机制构建面向不同轨迹长度的多个预测模型,以提高模型的训练效果.最后,在3个公开真实数据集上进行实验,验证了所提方法相比于多种已有方法的优势.
Abstract
In the task of remaining time prediction of business process instances,only the prefixes of process instances are encoded in most methods.The impact of events that have occurred on prediction results can only be considered,and the impact of the future events is ignored in such prediction methods.Aiming at this deficiency,a prediction framework for the remaining time of process instances was proposed,which integrated memory network and transition system.The suffixes of process instances were mined through transition system in this framework,then the prefixes and suffixes of the process were encoded based on the bidirectional gate recurrent unit.The memory network and attention mechanism were used to learn the weights of different events in process instances,so as to better model process instances.In addition,to fully consider the correlation and quantity difference between process instance prefixes of different lengths,the transfer-learning mechanism was introduced to construct multiple predic-tion models for different track lengths to improve the pertinence of the model.Finally,experiments were conducted on three public real-world datasets,and the results showed that the proposed method had obvious advantages over a variety of existing methods.
关键词
业务过程/剩余执行时间/变迁系统/深度学习Key words
business process/remaining execution time/transition system/deep learning引用本文复制引用
基金项目
山东省自然科学基金资助项目(ZR2022MF319)
国家自然科学基金资助项目(71704096)
国家自然科学基金资助项目(U1931207)
青岛市社会科学规划研究项目(QDSKL1801122)
青岛市社会科学规划研究项目(QDSKL2001117)
山东省泰山学者工程专项基金资助项目(ts20190936)
出版年
2024
NETL
NSTL
万方数据